ip_state.c revision 153876
1/* $FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 153876 2005-12-30 11:32:23Z guido $ */ 2 3/* 4 * Copyright (C) 1995-2003 by Darren Reed. 5 * 6 * See the IPFILTER.LICENCE file for details on licencing. 7 */ 8#if defined(KERNEL) || defined(_KERNEL) 9# undef KERNEL 10# undef _KERNEL 11# define KERNEL 1 12# define _KERNEL 1 13#endif 14#include <sys/errno.h> 15#include <sys/types.h> 16#include <sys/param.h> 17#include <sys/file.h> 18#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \ 19 defined(_KERNEL) 20# include "opt_ipfilter_log.h" 21#endif 22#if defined(_KERNEL) && defined(__FreeBSD_version) && \ 23 (__FreeBSD_version >= 400000) && !defined(KLD_MODULE) 24#include "opt_inet6.h" 25#endif 26#if !defined(_KERNEL) && !defined(__KERNEL__) 27# include <stdio.h> 28# include <stdlib.h> 29# include <string.h> 30# define _KERNEL 31# ifdef __OpenBSD__ 32struct file; 33# endif 34# include <sys/uio.h> 35# undef _KERNEL 36#endif 37#if defined(_KERNEL) && (__FreeBSD_version >= 220000) 38# include <sys/filio.h> 39# include <sys/fcntl.h> 40# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM) 41# include "opt_ipfilter.h" 42# endif 43#else 44# include <sys/ioctl.h> 45#endif 46#include <sys/time.h> 47#if !defined(linux) 48# include <sys/protosw.h> 49#endif 50#include <sys/socket.h> 51#if defined(_KERNEL) 52# include <sys/systm.h> 53# if !defined(__SVR4) && !defined(__svr4__) 54# include <sys/mbuf.h> 55# endif 56#endif 57#if defined(__SVR4) || defined(__svr4__) 58# include <sys/filio.h> 59# include <sys/byteorder.h> 60# ifdef _KERNEL 61# include <sys/dditypes.h> 62# endif 63# include <sys/stream.h> 64# include <sys/kmem.h> 65#endif 66 67#include <net/if.h> 68#ifdef sun 69# include <net/af.h> 70#endif 71#include <net/route.h> 72#include <netinet/in.h> 73#include <netinet/in_systm.h> 74#include <netinet/ip.h> 75#include <netinet/tcp.h> 76#if !defined(linux) 77# include <netinet/ip_var.h> 78#endif 79#if !defined(__hpux) && !defined(linux) 80# include <netinet/tcp_fsm.h> 81#endif 82#include <netinet/udp.h> 83#include <netinet/ip_icmp.h> 84#include "netinet/ip_compat.h" 85#include <netinet/tcpip.h> 86#include "netinet/ip_fil.h" 87#include "netinet/ip_nat.h" 88#include "netinet/ip_frag.h" 89#include "netinet/ip_state.h" 90#include "netinet/ip_proxy.h" 91#ifdef IPFILTER_SYNC 92#include "netinet/ip_sync.h" 93#endif 94#ifdef IPFILTER_SCAN 95#include "netinet/ip_scan.h" 96#endif 97#ifdef USE_INET6 98#include <netinet/icmp6.h> 99#endif 100#if (__FreeBSD_version >= 300000) 101# include <sys/malloc.h> 102# if defined(_KERNEL) && !defined(IPFILTER_LKM) 103# include <sys/libkern.h> 104# include <sys/systm.h> 105# endif 106#endif 107/* END OF INCLUDES */ 108 109 110#if !defined(lint) 111static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed"; 112static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.186.2.36 2005/12/04 22:25:36 darrenr Exp $"; 113#endif 114 115static ipstate_t **ips_table = NULL; 116static u_long *ips_seed = NULL; 117static int ips_num = 0; 118static u_long ips_last_force_flush = 0; 119ips_stat_t ips_stats; 120 121#ifdef USE_INET6 122static ipstate_t *fr_checkicmp6matchingstate __P((fr_info_t *)); 123#endif 124static ipstate_t *fr_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *, 125 i6addr_t *, tcphdr_t *, u_32_t)); 126static ipstate_t *fr_checkicmpmatchingstate __P((fr_info_t *)); 127static int fr_state_flush __P((int, int)); 128static ips_stat_t *fr_statetstats __P((void)); 129static void fr_delstate __P((ipstate_t *, int)); 130static int fr_state_remove __P((caddr_t)); 131static void fr_ipsmove __P((ipstate_t *, u_int)); 132static int fr_tcpstate __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 133static int fr_tcpoptions __P((fr_info_t *, tcphdr_t *, tcpdata_t *)); 134static ipstate_t *fr_stclone __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 135static void fr_fixinisn __P((fr_info_t *, ipstate_t *)); 136static void fr_fixoutisn __P((fr_info_t *, ipstate_t *)); 137static void fr_checknewisn __P((fr_info_t *, ipstate_t *)); 138 139int fr_stputent __P((caddr_t)); 140int fr_stgetent __P((caddr_t)); 141 142#define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */ 143#define FIVE_DAYS (5 * ONE_DAY) 144#define DOUBLE_HASH(x) (((x) + ips_seed[(x) % fr_statesize]) % fr_statesize) 145 146u_long fr_tcpidletimeout = FIVE_DAYS, 147 fr_tcpclosewait = IPF_TTLVAL(2 * TCP_MSL), 148 fr_tcplastack = IPF_TTLVAL(2 * TCP_MSL), 149 fr_tcptimeout = IPF_TTLVAL(2 * TCP_MSL), 150 fr_tcpclosed = IPF_TTLVAL(60), 151 fr_tcphalfclosed = IPF_TTLVAL(2 * 3600), /* 2 hours */ 152 fr_udptimeout = IPF_TTLVAL(120), 153 fr_udpacktimeout = IPF_TTLVAL(12), 154 fr_icmptimeout = IPF_TTLVAL(60), 155 fr_icmpacktimeout = IPF_TTLVAL(6), 156 fr_iptimeout = IPF_TTLVAL(60); 157int fr_statemax = IPSTATE_MAX, 158 fr_statesize = IPSTATE_SIZE; 159int fr_state_doflush = 0, 160 fr_state_lock = 0, 161 fr_state_maxbucket = 0, 162 fr_state_maxbucket_reset = 1, 163 fr_state_init = 0; 164ipftq_t ips_tqtqb[IPF_TCP_NSTATES], 165 ips_udptq, 166 ips_udpacktq, 167 ips_iptq, 168 ips_icmptq, 169 ips_icmpacktq, 170 *ips_utqe = NULL; 171#ifdef IPFILTER_LOG 172int ipstate_logging = 1; 173#else 174int ipstate_logging = 0; 175#endif 176ipstate_t *ips_list = NULL; 177 178 179/* ------------------------------------------------------------------------ */ 180/* Function: fr_stateinit */ 181/* Returns: int - 0 == success, -1 == failure */ 182/* Parameters: Nil */ 183/* */ 184/* Initialise all the global variables used within the state code. */ 185/* This action also includes initiailising locks. */ 186/* ------------------------------------------------------------------------ */ 187int fr_stateinit() 188{ 189 int i; 190 191 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *)); 192 if (ips_table == NULL) 193 return -1; 194 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *)); 195 196 KMALLOCS(ips_seed, u_long *, fr_statesize * sizeof(*ips_seed)); 197 if (ips_seed == NULL) 198 return -2; 199 for (i = 0; i < fr_statesize; i++) { 200 /* 201 * XXX - ips_seed[X] should be a random number of sorts. 202 */ 203#if (__FreeBSD_version >= 400000) 204 ips_seed[i] = arc4random(); 205#else 206 ips_seed[i] = ((u_long)ips_seed + i) * fr_statesize; 207 ips_seed[i] ^= 0xa5a55a5a; 208 ips_seed[i] *= (u_long)ips_seed; 209 ips_seed[i] ^= 0x5a5aa5a5; 210 ips_seed[i] *= fr_statemax; 211#endif 212 } 213 214 /* fill icmp reply type table */ 215 for (i = 0; i <= ICMP_MAXTYPE; i++) 216 icmpreplytype4[i] = -1; 217 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 218 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 219 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 220 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 221#ifdef USE_INET6 222 /* fill icmp reply type table */ 223 for (i = 0; i <= ICMP6_MAXTYPE; i++) 224 icmpreplytype6[i] = -1; 225 icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY; 226 icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT; 227 icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY; 228 icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT; 229 icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT; 230#endif 231 232 KMALLOCS(ips_stats.iss_bucketlen, u_long *, 233 fr_statesize * sizeof(u_long)); 234 if (ips_stats.iss_bucketlen == NULL) 235 return -1; 236 bzero((char *)ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long)); 237 238 if (fr_state_maxbucket == 0) { 239 for (i = fr_statesize; i > 0; i >>= 1) 240 fr_state_maxbucket++; 241 fr_state_maxbucket *= 2; 242 } 243 244 fr_sttab_init(ips_tqtqb); 245 ips_tqtqb[IPF_TCP_NSTATES - 1].ifq_next = &ips_udptq; 246 ips_udptq.ifq_ttl = (u_long)fr_udptimeout; 247 ips_udptq.ifq_ref = 1; 248 ips_udptq.ifq_head = NULL; 249 ips_udptq.ifq_tail = &ips_udptq.ifq_head; 250 MUTEX_INIT(&ips_udptq.ifq_lock, "ipftq udp tab"); 251 ips_udptq.ifq_next = &ips_udpacktq; 252 ips_udpacktq.ifq_ttl = (u_long)fr_udpacktimeout; 253 ips_udpacktq.ifq_ref = 1; 254 ips_udpacktq.ifq_head = NULL; 255 ips_udpacktq.ifq_tail = &ips_udpacktq.ifq_head; 256 MUTEX_INIT(&ips_udpacktq.ifq_lock, "ipftq udpack tab"); 257 ips_udpacktq.ifq_next = &ips_icmptq; 258 ips_icmptq.ifq_ttl = (u_long)fr_icmptimeout; 259 ips_icmptq.ifq_ref = 1; 260 ips_icmptq.ifq_head = NULL; 261 ips_icmptq.ifq_tail = &ips_icmptq.ifq_head; 262 MUTEX_INIT(&ips_icmptq.ifq_lock, "ipftq icmp tab"); 263 ips_icmptq.ifq_next = &ips_icmpacktq; 264 ips_icmpacktq.ifq_ttl = (u_long)fr_icmpacktimeout; 265 ips_icmpacktq.ifq_ref = 1; 266 ips_icmpacktq.ifq_head = NULL; 267 ips_icmpacktq.ifq_tail = &ips_icmpacktq.ifq_head; 268 MUTEX_INIT(&ips_icmpacktq.ifq_lock, "ipftq icmpack tab"); 269 ips_icmpacktq.ifq_next = &ips_iptq; 270 ips_iptq.ifq_ttl = (u_long)fr_iptimeout; 271 ips_iptq.ifq_ref = 1; 272 ips_iptq.ifq_head = NULL; 273 ips_iptq.ifq_tail = &ips_iptq.ifq_head; 274 MUTEX_INIT(&ips_iptq.ifq_lock, "ipftq ip tab"); 275 ips_iptq.ifq_next = NULL; 276 277 RWLOCK_INIT(&ipf_state, "ipf IP state rwlock"); 278 MUTEX_INIT(&ipf_stinsert, "ipf state insert mutex"); 279 fr_state_init = 1; 280 281 ips_last_force_flush = fr_ticks; 282 return 0; 283} 284 285 286/* ------------------------------------------------------------------------ */ 287/* Function: fr_stateunload */ 288/* Returns: Nil */ 289/* Parameters: Nil */ 290/* */ 291/* Release and destroy any resources acquired or initialised so that */ 292/* IPFilter can be unloaded or re-initialised. */ 293/* ------------------------------------------------------------------------ */ 294void fr_stateunload() 295{ 296 ipftq_t *ifq, *ifqnext; 297 ipstate_t *is; 298 299 while ((is = ips_list) != NULL) 300 fr_delstate(is, 0); 301 302 /* 303 * Proxy timeout queues are not cleaned here because although they 304 * exist on the state list, appr_unload is called after fr_stateunload 305 * and the proxies actually are responsible for them being created. 306 * Should the proxy timeouts have their own list? There's no real 307 * justification as this is the only complicationA 308 */ 309 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 310 ifqnext = ifq->ifq_next; 311 if (((ifq->ifq_flags & IFQF_PROXY) == 0) && 312 (fr_deletetimeoutqueue(ifq) == 0)) 313 fr_freetimeoutqueue(ifq); 314 } 315 316 ips_stats.iss_inuse = 0; 317 ips_num = 0; 318 319 if (fr_state_init == 1) { 320 fr_sttab_destroy(ips_tqtqb); 321 MUTEX_DESTROY(&ips_udptq.ifq_lock); 322 MUTEX_DESTROY(&ips_icmptq.ifq_lock); 323 MUTEX_DESTROY(&ips_udpacktq.ifq_lock); 324 MUTEX_DESTROY(&ips_icmpacktq.ifq_lock); 325 MUTEX_DESTROY(&ips_iptq.ifq_lock); 326 } 327 328 if (ips_table != NULL) { 329 KFREES(ips_table, fr_statesize * sizeof(*ips_table)); 330 ips_table = NULL; 331 } 332 333 if (ips_seed != NULL) { 334 KFREES(ips_seed, fr_statesize * sizeof(*ips_seed)); 335 ips_seed = NULL; 336 } 337 338 if (ips_stats.iss_bucketlen != NULL) { 339 KFREES(ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long)); 340 ips_stats.iss_bucketlen = NULL; 341 } 342 343 if (fr_state_maxbucket_reset == 1) 344 fr_state_maxbucket = 0; 345 346 if (fr_state_init == 1) { 347 fr_state_init = 0; 348 RW_DESTROY(&ipf_state); 349 MUTEX_DESTROY(&ipf_stinsert); 350 } 351} 352 353 354/* ------------------------------------------------------------------------ */ 355/* Function: fr_statetstats */ 356/* Returns: ips_state_t* - pointer to state stats structure */ 357/* Parameters: Nil */ 358/* */ 359/* Put all the current numbers and pointers into a single struct and return */ 360/* a pointer to it. */ 361/* ------------------------------------------------------------------------ */ 362static ips_stat_t *fr_statetstats() 363{ 364 ips_stats.iss_active = ips_num; 365 ips_stats.iss_statesize = fr_statesize; 366 ips_stats.iss_statemax = fr_statemax; 367 ips_stats.iss_table = ips_table; 368 ips_stats.iss_list = ips_list; 369 ips_stats.iss_ticks = fr_ticks; 370 return &ips_stats; 371} 372 373/* ------------------------------------------------------------------------ */ 374/* Function: fr_state_remove */ 375/* Returns: int - 0 == success, != 0 == failure */ 376/* Parameters: data(I) - pointer to state structure to delete from table */ 377/* */ 378/* Search for a state structure that matches the one passed, according to */ 379/* the IP addresses and other protocol specific information. */ 380/* ------------------------------------------------------------------------ */ 381static int fr_state_remove(data) 382caddr_t data; 383{ 384 ipstate_t *sp, st; 385 int error; 386 387 sp = &st; 388 error = fr_inobj(data, &st, IPFOBJ_IPSTATE); 389 if (error) 390 return EFAULT; 391 392 WRITE_ENTER(&ipf_state); 393 for (sp = ips_list; sp; sp = sp->is_next) 394 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 395 !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src, 396 sizeof(st.is_src)) && 397 !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_src, 398 sizeof(st.is_dst)) && 399 !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps, 400 sizeof(st.is_ps))) { 401 fr_delstate(sp, ISL_REMOVE); 402 RWLOCK_EXIT(&ipf_state); 403 return 0; 404 } 405 RWLOCK_EXIT(&ipf_state); 406 return ESRCH; 407} 408 409 410/* ------------------------------------------------------------------------ */ 411/* Function: fr_state_ioctl */ 412/* Returns: int - 0 == success, != 0 == failure */ 413/* Parameters: data(I) - pointer to ioctl data */ 414/* cmd(I) - ioctl command integer */ 415/* mode(I) - file mode bits used with open */ 416/* */ 417/* Processes an ioctl call made to operate on the IP Filter state device. */ 418/* ------------------------------------------------------------------------ */ 419int fr_state_ioctl(data, cmd, mode) 420caddr_t data; 421ioctlcmd_t cmd; 422int mode; 423{ 424 int arg, ret, error = 0; 425 426 switch (cmd) 427 { 428 /* 429 * Delete an entry from the state table. 430 */ 431 case SIOCDELST : 432 error = fr_state_remove(data); 433 break; 434 /* 435 * Flush the state table 436 */ 437 case SIOCIPFFL : 438 BCOPYIN(data, (char *)&arg, sizeof(arg)); 439 if (arg == 0 || arg == 1) { 440 WRITE_ENTER(&ipf_state); 441 ret = fr_state_flush(arg, 4); 442 RWLOCK_EXIT(&ipf_state); 443 BCOPYOUT((char *)&ret, data, sizeof(ret)); 444 } else 445 error = EINVAL; 446 break; 447#ifdef USE_INET6 448 case SIOCIPFL6 : 449 BCOPYIN(data, (char *)&arg, sizeof(arg)); 450 if (arg == 0 || arg == 1) { 451 WRITE_ENTER(&ipf_state); 452 ret = fr_state_flush(arg, 6); 453 RWLOCK_EXIT(&ipf_state); 454 BCOPYOUT((char *)&ret, data, sizeof(ret)); 455 } else 456 error = EINVAL; 457 break; 458#endif 459#ifdef IPFILTER_LOG 460 /* 461 * Flush the state log. 462 */ 463 case SIOCIPFFB : 464 if (!(mode & FWRITE)) 465 error = EPERM; 466 else { 467 int tmp; 468 469 tmp = ipflog_clear(IPL_LOGSTATE); 470 BCOPYOUT((char *)&tmp, data, sizeof(tmp)); 471 } 472 break; 473 /* 474 * Turn logging of state information on/off. 475 */ 476 case SIOCSETLG : 477 if (!(mode & FWRITE)) 478 error = EPERM; 479 else { 480 BCOPYIN((char *)data, (char *)&ipstate_logging, 481 sizeof(ipstate_logging)); 482 } 483 break; 484 /* 485 * Return the current state of logging. 486 */ 487 case SIOCGETLG : 488 BCOPYOUT((char *)&ipstate_logging, (char *)data, 489 sizeof(ipstate_logging)); 490 break; 491 /* 492 * Return the number of bytes currently waiting to be read. 493 */ 494 case FIONREAD : 495 arg = iplused[IPL_LOGSTATE]; /* returned in an int */ 496 BCOPYOUT((char *)&arg, data, sizeof(arg)); 497 break; 498#endif 499 /* 500 * Get the current state statistics. 501 */ 502 case SIOCGETFS : 503 error = fr_outobj(data, fr_statetstats(), IPFOBJ_STATESTAT); 504 break; 505 /* 506 * Lock/Unlock the state table. (Locking prevents any changes, which 507 * means no packets match). 508 */ 509 case SIOCSTLCK : 510 if (!(mode & FWRITE)) { 511 error = EPERM; 512 } else { 513 fr_lock(data, &fr_state_lock); 514 } 515 break; 516 /* 517 * Add an entry to the current state table. 518 */ 519 case SIOCSTPUT : 520 if (!fr_state_lock || !(mode &FWRITE)) { 521 error = EACCES; 522 break; 523 } 524 error = fr_stputent(data); 525 break; 526 /* 527 * Get a state table entry. 528 */ 529 case SIOCSTGET : 530 if (!fr_state_lock) { 531 error = EACCES; 532 break; 533 } 534 error = fr_stgetent(data); 535 break; 536 default : 537 error = EINVAL; 538 break; 539 } 540 return error; 541} 542 543 544/* ------------------------------------------------------------------------ */ 545/* Function: fr_stgetent */ 546/* Returns: int - 0 == success, != 0 == failure */ 547/* Parameters: data(I) - pointer to state structure to retrieve from table */ 548/* */ 549/* Copy out state information from the kernel to a user space process. If */ 550/* there is a filter rule associated with the state entry, copy that out */ 551/* as well. The entry to copy out is taken from the value of "ips_next" in */ 552/* the struct passed in and if not null and not found in the list of current*/ 553/* state entries, the retrieval fails. */ 554/* ------------------------------------------------------------------------ */ 555int fr_stgetent(data) 556caddr_t data; 557{ 558 ipstate_t *is, *isn; 559 ipstate_save_t ips; 560 int error; 561 562 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 563 if (error) 564 return EFAULT; 565 566 isn = ips.ips_next; 567 if (isn == NULL) { 568 isn = ips_list; 569 if (isn == NULL) { 570 if (ips.ips_next == NULL) 571 return ENOENT; 572 return 0; 573 } 574 } else { 575 /* 576 * Make sure the pointer we're copying from exists in the 577 * current list of entries. Security precaution to prevent 578 * copying of random kernel data. 579 */ 580 for (is = ips_list; is; is = is->is_next) 581 if (is == isn) 582 break; 583 if (!is) 584 return ESRCH; 585 } 586 ips.ips_next = isn->is_next; 587 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 588 ips.ips_rule = isn->is_rule; 589 if (isn->is_rule != NULL) 590 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 591 sizeof(ips.ips_fr)); 592 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 593 if (error) 594 return EFAULT; 595 return 0; 596} 597 598 599/* ------------------------------------------------------------------------ */ 600/* Function: fr_stputent */ 601/* Returns: int - 0 == success, != 0 == failure */ 602/* Parameters: data(I) - pointer to state information struct */ 603/* */ 604/* This function implements the SIOCSTPUT ioctl: insert a state entry into */ 605/* the state table. If the state info. includes a pointer to a filter rule */ 606/* then also add in an orphaned rule (will not show up in any "ipfstat -io" */ 607/* output. */ 608/* ------------------------------------------------------------------------ */ 609int fr_stputent(data) 610caddr_t data; 611{ 612 ipstate_t *is, *isn; 613 ipstate_save_t ips; 614 int error, out, i; 615 frentry_t *fr; 616 char *name; 617 618 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 619 if (error) 620 return EFAULT; 621 622 KMALLOC(isn, ipstate_t *); 623 if (isn == NULL) 624 return ENOMEM; 625 626 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 627 bzero((char *)isn, offsetof(struct ipstate, is_pkts)); 628 isn->is_sti.tqe_pnext = NULL; 629 isn->is_sti.tqe_next = NULL; 630 isn->is_sti.tqe_ifq = NULL; 631 isn->is_sti.tqe_parent = isn; 632 isn->is_ifp[0] = NULL; 633 isn->is_ifp[1] = NULL; 634 isn->is_ifp[2] = NULL; 635 isn->is_ifp[3] = NULL; 636 isn->is_sync = NULL; 637 fr = ips.ips_rule; 638 639 if (fr == NULL) { 640 READ_ENTER(&ipf_state); 641 fr_stinsert(isn, 0); 642 MUTEX_EXIT(&isn->is_lock); 643 RWLOCK_EXIT(&ipf_state); 644 return 0; 645 } 646 647 if (isn->is_flags & SI_NEWFR) { 648 KMALLOC(fr, frentry_t *); 649 if (fr == NULL) { 650 KFREE(isn); 651 return ENOMEM; 652 } 653 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 654 out = fr->fr_flags & FR_OUTQUE ? 1 : 0; 655 isn->is_rule = fr; 656 ips.ips_is.is_rule = fr; 657 MUTEX_NUKE(&fr->fr_lock); 658 MUTEX_INIT(&fr->fr_lock, "state filter rule lock"); 659 660 /* 661 * Look up all the interface names in the rule. 662 */ 663 for (i = 0; i < 4; i++) { 664 name = fr->fr_ifnames[i]; 665 fr->fr_ifas[i] = fr_resolvenic(name, fr->fr_v); 666 name = isn->is_ifname[i]; 667 isn->is_ifp[i] = fr_resolvenic(name, isn->is_v); 668 } 669 670 fr->fr_ref = 0; 671 fr->fr_dsize = 0; 672 fr->fr_data = NULL; 673 674 fr_resolvedest(&fr->fr_tif, fr->fr_v); 675 fr_resolvedest(&fr->fr_dif, fr->fr_v); 676 677 /* 678 * send a copy back to userland of what we ended up 679 * to allow for verification. 680 */ 681 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 682 if (error) { 683 KFREE(isn); 684 MUTEX_DESTROY(&fr->fr_lock); 685 KFREE(fr); 686 return EFAULT; 687 } 688 READ_ENTER(&ipf_state); 689 fr_stinsert(isn, 0); 690 MUTEX_EXIT(&isn->is_lock); 691 RWLOCK_EXIT(&ipf_state); 692 693 } else { 694 READ_ENTER(&ipf_state); 695 for (is = ips_list; is; is = is->is_next) 696 if (is->is_rule == fr) { 697 fr_stinsert(isn, 0); 698 MUTEX_EXIT(&isn->is_lock); 699 break; 700 } 701 702 if (is == NULL) { 703 KFREE(isn); 704 isn = NULL; 705 } 706 RWLOCK_EXIT(&ipf_state); 707 708 return (isn == NULL) ? ESRCH : 0; 709 } 710 711 return 0; 712} 713 714 715/* ------------------------------------------------------------------------ */ 716/* Function: fr_stinsert */ 717/* Returns: Nil */ 718/* Parameters: is(I) - pointer to state structure */ 719/* rev(I) - flag indicating forward/reverse direction of packet */ 720/* */ 721/* Inserts a state structure into the hash table (for lookups) and the list */ 722/* of state entries (for enumeration). Resolves all of the interface names */ 723/* to pointers and adjusts running stats for the hash table as appropriate. */ 724/* */ 725/* Locking: it is assumed that some kind of lock on ipf_state is held. */ 726/* Exits with is_lock initialised and held. */ 727/* ------------------------------------------------------------------------ */ 728void fr_stinsert(is, rev) 729ipstate_t *is; 730int rev; 731{ 732 frentry_t *fr; 733 u_int hv; 734 int i; 735 736 MUTEX_INIT(&is->is_lock, "ipf state entry"); 737 738 fr = is->is_rule; 739 if (fr != NULL) { 740 MUTEX_ENTER(&fr->fr_lock); 741 fr->fr_ref++; 742 fr->fr_statecnt++; 743 MUTEX_EXIT(&fr->fr_lock); 744 } 745 746 /* 747 * Look up all the interface names in the state entry. 748 */ 749 for (i = 0; i < 4; i++) { 750 if (is->is_ifp[i] != NULL) 751 continue; 752 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], is->is_v); 753 } 754 755 /* 756 * If we could trust is_hv, then the modulous would not be needed, but 757 * when running with IPFILTER_SYNC, this stops bad values. 758 */ 759 hv = is->is_hv % fr_statesize; 760 is->is_hv = hv; 761 762 /* 763 * We need to get both of these locks...the first because it is 764 * possible that once the insert is complete another packet might 765 * come along, match the entry and want to update it. 766 */ 767 MUTEX_ENTER(&is->is_lock); 768 MUTEX_ENTER(&ipf_stinsert); 769 770 /* 771 * add into list table. 772 */ 773 if (ips_list != NULL) 774 ips_list->is_pnext = &is->is_next; 775 is->is_pnext = &ips_list; 776 is->is_next = ips_list; 777 ips_list = is; 778 779 if (ips_table[hv] != NULL) 780 ips_table[hv]->is_phnext = &is->is_hnext; 781 else 782 ips_stats.iss_inuse++; 783 is->is_phnext = ips_table + hv; 784 is->is_hnext = ips_table[hv]; 785 ips_table[hv] = is; 786 ips_stats.iss_bucketlen[hv]++; 787 ips_num++; 788 MUTEX_EXIT(&ipf_stinsert); 789 790 fr_setstatequeue(is, rev); 791} 792 793 794/* ------------------------------------------------------------------------ */ 795/* Function: fr_addstate */ 796/* Returns: ipstate_t* - NULL == failure, else pointer to new state */ 797/* Parameters: fin(I) - pointer to packet information */ 798/* stsave(O) - pointer to place to save pointer to created */ 799/* state structure. */ 800/* flags(I) - flags to use when creating the structure */ 801/* */ 802/* Creates a new IP state structure from the packet information collected. */ 803/* Inserts it into the state table and appends to the bottom of the active */ 804/* list. If the capacity of the table has reached the maximum allowed then */ 805/* the call will fail and a flush is scheduled for the next timeout call. */ 806/* ------------------------------------------------------------------------ */ 807ipstate_t *fr_addstate(fin, stsave, flags) 808fr_info_t *fin; 809ipstate_t **stsave; 810u_int flags; 811{ 812 ipstate_t *is, ips; 813 struct icmp *ic; 814 u_int pass, hv; 815 frentry_t *fr; 816 tcphdr_t *tcp; 817 grehdr_t *gre; 818 void *ifp; 819 int out; 820 821 if (fr_state_lock || 822 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 823 return NULL; 824 825 if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN)) 826 return NULL; 827 828 fr = fin->fin_fr; 829 if ((fr->fr_statemax == 0) && (ips_num == fr_statemax)) { 830 ATOMIC_INCL(ips_stats.iss_max); 831 fr_state_doflush = 1; 832 return NULL; 833 } 834 835 /* 836 * If a "keep state" rule has reached the maximum number of references 837 * to it, then schedule an automatic flush in case we can clear out 838 * some "dead old wood". 839 */ 840 MUTEX_ENTER(&fr->fr_lock); 841 if ((fr != NULL) && (fr->fr_statemax != 0) && 842 (fr->fr_statecnt >= fr->fr_statemax)) { 843 MUTEX_EXIT(&fr->fr_lock); 844 ATOMIC_INCL(ips_stats.iss_maxref); 845 fr_state_doflush = 1; 846 return NULL; 847 } 848 fr->fr_statecnt++; 849 MUTEX_EXIT(&fr->fr_lock); 850 851 pass = (fr == NULL) ? 0 : fr->fr_flags; 852 853 ic = NULL; 854 tcp = NULL; 855 out = fin->fin_out; 856 is = &ips; 857 bzero((char *)is, sizeof(*is)); 858 is->is_die = 1 + fr_ticks; 859 860 /* 861 * Copy and calculate... 862 */ 863 hv = (is->is_p = fin->fin_fi.fi_p); 864 is->is_src = fin->fin_fi.fi_src; 865 hv += is->is_saddr; 866 is->is_dst = fin->fin_fi.fi_dst; 867 hv += is->is_daddr; 868#ifdef USE_INET6 869 if (fin->fin_v == 6) { 870 /* 871 * For ICMPv6, we check to see if the destination address is 872 * a multicast address. If it is, do not include it in the 873 * calculation of the hash because the correct reply will come 874 * back from a real address, not a multicast address. 875 */ 876 if ((is->is_p == IPPROTO_ICMPV6) && 877 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) { 878 /* 879 * So you can do keep state with neighbour discovery. 880 * 881 * Here we could use the address from the neighbour 882 * solicit message to put in the state structure and 883 * we could use that without a wildcard flag too... 884 */ 885 flags |= SI_W_DADDR; 886 hv -= is->is_daddr; 887 } else { 888 hv += is->is_dst.i6[1]; 889 hv += is->is_dst.i6[2]; 890 hv += is->is_dst.i6[3]; 891 } 892 hv += is->is_src.i6[1]; 893 hv += is->is_src.i6[2]; 894 hv += is->is_src.i6[3]; 895 } 896#endif 897 898 switch (is->is_p) 899 { 900#ifdef USE_INET6 901 case IPPROTO_ICMPV6 : 902 ic = fin->fin_dp; 903 904 switch (ic->icmp_type) 905 { 906 case ICMP6_ECHO_REQUEST : 907 is->is_icmp.ici_type = ic->icmp_type; 908 hv += (is->is_icmp.ici_id = ic->icmp_id); 909 break; 910 case ICMP6_MEMBERSHIP_QUERY : 911 case ND_ROUTER_SOLICIT : 912 case ND_NEIGHBOR_SOLICIT : 913 case ICMP6_NI_QUERY : 914 is->is_icmp.ici_type = ic->icmp_type; 915 break; 916 default : 917 return NULL; 918 } 919 ATOMIC_INCL(ips_stats.iss_icmp); 920 break; 921#endif 922 case IPPROTO_ICMP : 923 ic = fin->fin_dp; 924 925 switch (ic->icmp_type) 926 { 927 case ICMP_ECHO : 928 case ICMP_TSTAMP : 929 case ICMP_IREQ : 930 case ICMP_MASKREQ : 931 is->is_icmp.ici_type = ic->icmp_type; 932 hv += (is->is_icmp.ici_id = ic->icmp_id); 933 break; 934 default : 935 return NULL; 936 } 937 ATOMIC_INCL(ips_stats.iss_icmp); 938 break; 939 940 case IPPROTO_GRE : 941 gre = fin->fin_dp; 942 943 is->is_gre.gs_flags = gre->gr_flags; 944 is->is_gre.gs_ptype = gre->gr_ptype; 945 if (GRE_REV(is->is_gre.gs_flags) == 1) { 946 is->is_call[0] = fin->fin_data[0]; 947 is->is_call[1] = fin->fin_data[1]; 948 } 949 break; 950 951 case IPPROTO_TCP : 952 tcp = fin->fin_dp; 953 954 if (tcp->th_flags & TH_RST) 955 return NULL; 956 /* 957 * The endian of the ports doesn't matter, but the ack and 958 * sequence numbers do as we do mathematics on them later. 959 */ 960 is->is_sport = htons(fin->fin_data[0]); 961 is->is_dport = htons(fin->fin_data[1]); 962 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 963 hv += is->is_sport; 964 hv += is->is_dport; 965 } 966 967 /* 968 * If this is a real packet then initialise fields in the 969 * state information structure from the TCP header information. 970 */ 971 972 is->is_maxdwin = 1; 973 is->is_maxswin = ntohs(tcp->th_win); 974 if (is->is_maxswin == 0) 975 is->is_maxswin = 1; 976 977 if ((fin->fin_flx & FI_IGNORE) == 0) { 978 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 979 (TCP_OFF(tcp) << 2) + 980 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 981 ((tcp->th_flags & TH_FIN) ? 1 : 0); 982 is->is_maxsend = is->is_send; 983 984 /* 985 * Window scale option is only present in 986 * SYN/SYN-ACK packet. 987 */ 988 if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) == 989 TH_SYN && 990 (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 991 if (fr_tcpoptions(fin, tcp, 992 &is->is_tcp.ts_data[0]) == -1) { 993 fin->fin_flx |= FI_BAD; 994 } 995 } 996 997 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) { 998 fr_checknewisn(fin, is); 999 fr_fixoutisn(fin, is); 1000 } 1001 1002 if ((tcp->th_flags & TH_OPENING) == TH_SYN) 1003 flags |= IS_TCPFSM; 1004 else { 1005 is->is_maxdwin = is->is_maxswin * 2; 1006 is->is_dend = ntohl(tcp->th_ack); 1007 is->is_maxdend = ntohl(tcp->th_ack); 1008 is->is_maxdwin *= 2; 1009 } 1010 } 1011 1012 /* 1013 * If we're creating state for a starting connection, start the 1014 * timer on it as we'll never see an error if it fails to 1015 * connect. 1016 */ 1017 ATOMIC_INCL(ips_stats.iss_tcp); 1018 break; 1019 1020 case IPPROTO_UDP : 1021 tcp = fin->fin_dp; 1022 1023 is->is_sport = htons(fin->fin_data[0]); 1024 is->is_dport = htons(fin->fin_data[1]); 1025 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 1026 hv += tcp->th_dport; 1027 hv += tcp->th_sport; 1028 } 1029 ATOMIC_INCL(ips_stats.iss_udp); 1030 break; 1031 1032 default : 1033 break; 1034 } 1035 hv = DOUBLE_HASH(hv); 1036 is->is_hv = hv; 1037 is->is_rule = fr; 1038 is->is_flags = flags & IS_INHERITED; 1039 1040 /* 1041 * Look for identical state. 1042 */ 1043 for (is = ips_table[is->is_hv % fr_statesize]; is != NULL; 1044 is = is->is_hnext) { 1045 if (bcmp(&ips.is_src, &is->is_src, 1046 offsetof(struct ipstate, is_ps) - 1047 offsetof(struct ipstate, is_src)) == 0) 1048 break; 1049 } 1050 if (is != NULL) 1051 goto cantaddstate; 1052 1053 if (ips_stats.iss_bucketlen[hv] >= fr_state_maxbucket) { 1054 ATOMIC_INCL(ips_stats.iss_bucketfull); 1055 goto cantaddstate; 1056 } 1057 KMALLOC(is, ipstate_t *); 1058 if (is == NULL) { 1059 ATOMIC_INCL(ips_stats.iss_nomem); 1060 goto cantaddstate; 1061 } 1062 bcopy((char *)&ips, (char *)is, sizeof(*is)); 1063 /* 1064 * Do not do the modulous here, it is done in fr_stinsert(). 1065 */ 1066 if (fr != NULL) { 1067 (void) strncpy(is->is_group, fr->fr_group, FR_GROUPLEN); 1068 if (fr->fr_age[0] != 0) { 1069 is->is_tqehead[0] = fr_addtimeoutqueue(&ips_utqe, 1070 fr->fr_age[0]); 1071 is->is_sti.tqe_flags |= TQE_RULEBASED; 1072 } 1073 if (fr->fr_age[1] != 0) { 1074 is->is_tqehead[1] = fr_addtimeoutqueue(&ips_utqe, 1075 fr->fr_age[1]); 1076 is->is_sti.tqe_flags |= TQE_RULEBASED; 1077 } 1078 1079 is->is_tag = fr->fr_logtag; 1080 1081 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1]; 1082 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2]; 1083 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3]; 1084 1085 if (((ifp = fr->fr_ifas[1]) != NULL) && 1086 (ifp != (void *)-1)) { 1087 COPYIFNAME(ifp, is->is_ifname[(out << 1) + 1]); 1088 } 1089 if (((ifp = fr->fr_ifas[2]) != NULL) && 1090 (ifp != (void *)-1)) { 1091 COPYIFNAME(ifp, is->is_ifname[(1 - out) << 1]); 1092 } 1093 if (((ifp = fr->fr_ifas[3]) != NULL) && 1094 (ifp != (void *)-1)) { 1095 COPYIFNAME(ifp, is->is_ifname[((1 - out) << 1) + 1]); 1096 } 1097 } else { 1098 pass = fr_flags; 1099 is->is_tag = FR_NOLOGTAG; 1100 } 1101 1102 is->is_ifp[out << 1] = fin->fin_ifp; 1103 if (fin->fin_ifp != NULL) { 1104 COPYIFNAME(fin->fin_ifp, is->is_ifname[out << 1]); 1105 } 1106 1107 /* 1108 * It may seem strange to set is_ref to 2, but fr_check() will call 1109 * fr_statederef() after calling fr_addstate() and the idea is to 1110 * have it exist at the end of fr_check() with is_ref == 1. 1111 */ 1112 is->is_ref = 2; 1113 is->is_pass = pass; 1114 is->is_pkts[0] = 0, is->is_bytes[0] = 0; 1115 is->is_pkts[1] = 0, is->is_bytes[1] = 0; 1116 is->is_pkts[2] = 0, is->is_bytes[2] = 0; 1117 is->is_pkts[3] = 0, is->is_bytes[3] = 0; 1118 if ((fin->fin_flx & FI_IGNORE) == 0) { 1119 is->is_pkts[out] = 1; 1120 is->is_bytes[out] = fin->fin_plen; 1121 is->is_flx[out][0] = fin->fin_flx & FI_CMP; 1122 is->is_flx[out][0] &= ~FI_OOW; 1123 } 1124 1125 if (pass & FR_STSTRICT) 1126 is->is_flags |= IS_STRICT; 1127 1128 if (pass & FR_STATESYNC) 1129 is->is_flags |= IS_STATESYNC; 1130 1131 /* 1132 * We want to check everything that is a property of this packet, 1133 * but we don't (automatically) care about it's fragment status as 1134 * this may change. 1135 */ 1136 is->is_v = fin->fin_v; 1137 is->is_opt[0] = fin->fin_optmsk; 1138 is->is_optmsk[0] = 0xffffffff; 1139 is->is_optmsk[1] = 0xffffffff; 1140 if (is->is_v == 6) { 1141 is->is_opt[0] &= ~0x8; 1142 is->is_optmsk[0] &= ~0x8; 1143 is->is_optmsk[1] &= ~0x8; 1144 } 1145 is->is_sec = fin->fin_secmsk; 1146 is->is_secmsk = 0xffff; 1147 is->is_auth = fin->fin_auth; 1148 is->is_authmsk = 0xffff; 1149 if (flags & (SI_WILDP|SI_WILDA)) { 1150 ATOMIC_INCL(ips_stats.iss_wild); 1151 } 1152 is->is_rulen = fin->fin_rule; 1153 1154 1155 if (pass & FR_LOGFIRST) 1156 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 1157 1158 READ_ENTER(&ipf_state); 1159 is->is_me = stsave; 1160 1161 fr_stinsert(is, fin->fin_rev); 1162 1163 if (fin->fin_p == IPPROTO_TCP) { 1164 /* 1165 * If we're creating state for a starting connection, start the 1166 * timer on it as we'll never see an error if it fails to 1167 * connect. 1168 */ 1169 (void) fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags); 1170 MUTEX_EXIT(&is->is_lock); 1171#ifdef IPFILTER_SCAN 1172 if ((is->is_flags & SI_CLONE) == 0) 1173 (void) ipsc_attachis(is); 1174#endif 1175 } else { 1176 MUTEX_EXIT(&is->is_lock); 1177 } 1178#ifdef IPFILTER_SYNC 1179 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0)) 1180 is->is_sync = ipfsync_new(SMC_STATE, fin, is); 1181#endif 1182 if (ipstate_logging) 1183 ipstate_log(is, ISL_NEW); 1184 1185 RWLOCK_EXIT(&ipf_state); 1186 fin->fin_state = is; 1187 fin->fin_rev = IP6_NEQ(&is->is_dst, &fin->fin_daddr); 1188 fin->fin_flx |= FI_STATE; 1189 if (fin->fin_flx & FI_FRAG) 1190 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 1191 1192 return is; 1193 1194cantaddstate: 1195 if (fr != NULL) { 1196 MUTEX_ENTER(&fr->fr_lock); 1197 fr->fr_statecnt--; 1198 MUTEX_EXIT(&fr->fr_lock); 1199 } 1200 return NULL; 1201} 1202 1203 1204/* ------------------------------------------------------------------------ */ 1205/* Function: fr_tcpoptions */ 1206/* Returns: int - 1 == packet matches state entry, 0 == it does not, */ 1207/* -1 == packet has bad TCP options data */ 1208/* Parameters: fin(I) - pointer to packet information */ 1209/* tcp(I) - pointer to TCP packet header */ 1210/* td(I) - pointer to TCP data held as part of the state */ 1211/* */ 1212/* Look after the TCP header for any options and deal with those that are */ 1213/* present. Record details about those that we recogise. */ 1214/* ------------------------------------------------------------------------ */ 1215static int fr_tcpoptions(fin, tcp, td) 1216fr_info_t *fin; 1217tcphdr_t *tcp; 1218tcpdata_t *td; 1219{ 1220 int off, mlen, ol, i, len, retval; 1221 char buf[64], *s, opt; 1222 mb_t *m = NULL; 1223 1224 len = (TCP_OFF(tcp) << 2); 1225 if (fin->fin_dlen < len) 1226 return 0; 1227 len -= sizeof(*tcp); 1228 1229 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff; 1230 1231 m = fin->fin_m; 1232 mlen = MSGDSIZE(m) - off; 1233 if (len > mlen) { 1234 len = mlen; 1235 retval = 0; 1236 } else { 1237 retval = 1; 1238 } 1239 1240 COPYDATA(m, off, len, buf); 1241 1242 for (s = buf; len > 0; ) { 1243 opt = *s; 1244 if (opt == TCPOPT_EOL) 1245 break; 1246 else if (opt == TCPOPT_NOP) 1247 ol = 1; 1248 else { 1249 if (len < 2) 1250 break; 1251 ol = (int)*(s + 1); 1252 if (ol < 2 || ol > len) 1253 break; 1254 1255 /* 1256 * Extract the TCP options we are interested in out of 1257 * the header and store them in the the tcpdata struct. 1258 */ 1259 switch (opt) 1260 { 1261 case TCPOPT_WINDOW : 1262 if (ol == TCPOLEN_WINDOW) { 1263 i = (int)*(s + 2); 1264 if (i > TCP_WSCALE_MAX) 1265 i = TCP_WSCALE_MAX; 1266 else if (i < 0) 1267 i = 0; 1268 td->td_winscale = i; 1269 td->td_winflags |= TCP_WSCALE_SEEN| 1270 TCP_WSCALE_FIRST; 1271 } else 1272 retval = -1; 1273 break; 1274 case TCPOPT_MAXSEG : 1275 /* 1276 * So, if we wanted to set the TCP MAXSEG, 1277 * it should be done here... 1278 */ 1279 if (ol == TCPOLEN_MAXSEG) { 1280 i = (int)*(s + 2); 1281 i <<= 8; 1282 i += (int)*(s + 3); 1283 td->td_maxseg = i; 1284 } else 1285 retval = -1; 1286 break; 1287 case TCPOPT_SACK_PERMITTED : 1288 if (ol == TCPOLEN_SACK_PERMITTED) 1289 td->td_winflags |= TCP_SACK_PERMIT; 1290 else 1291 retval = -1; 1292 break; 1293 } 1294 } 1295 len -= ol; 1296 s += ol; 1297 } 1298 return retval; 1299} 1300 1301 1302/* ------------------------------------------------------------------------ */ 1303/* Function: fr_tcpstate */ 1304/* Returns: int - 1 == packet matches state entry, 0 == it does not */ 1305/* Parameters: fin(I) - pointer to packet information */ 1306/* tcp(I) - pointer to TCP packet header */ 1307/* is(I) - pointer to master state structure */ 1308/* */ 1309/* Check to see if a packet with TCP headers fits within the TCP window. */ 1310/* Change timeout depending on whether new packet is a SYN-ACK returning */ 1311/* for a SYN or a RST or FIN which indicate time to close up shop. */ 1312/* ------------------------------------------------------------------------ */ 1313static int fr_tcpstate(fin, tcp, is) 1314fr_info_t *fin; 1315tcphdr_t *tcp; 1316ipstate_t *is; 1317{ 1318 int source, ret = 0, flags; 1319 tcpdata_t *fdata, *tdata; 1320 1321 source = !fin->fin_rev; 1322 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) && 1323 (ntohs(is->is_sport) != fin->fin_data[0])) 1324 source = 0; 1325 fdata = &is->is_tcp.ts_data[!source]; 1326 tdata = &is->is_tcp.ts_data[source]; 1327 1328 MUTEX_ENTER(&is->is_lock); 1329 if (fr_tcpinwindow(fin, fdata, tdata, tcp, is->is_flags)) { 1330#ifdef IPFILTER_SCAN 1331 if (is->is_flags & (IS_SC_CLIENT|IS_SC_SERVER)) { 1332 ipsc_packet(fin, is); 1333 if (FR_ISBLOCK(is->is_pass)) { 1334 MUTEX_EXIT(&is->is_lock); 1335 return 1; 1336 } 1337 } 1338#endif 1339 1340 /* 1341 * Nearing end of connection, start timeout. 1342 */ 1343 ret = fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags); 1344 if (ret == 0) { 1345 MUTEX_EXIT(&is->is_lock); 1346 return 0; 1347 } 1348 1349 /* 1350 * set s0's as appropriate. Use syn-ack packet as it 1351 * contains both pieces of required information. 1352 */ 1353 /* 1354 * Window scale option is only present in SYN/SYN-ACK packet. 1355 * Compare with ~TH_FIN to mask out T/TCP setups. 1356 */ 1357 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL); 1358 if (flags == (TH_SYN|TH_ACK)) { 1359 is->is_s0[source] = ntohl(tcp->th_ack); 1360 is->is_s0[!source] = ntohl(tcp->th_seq) + 1; 1361 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2)) && 1362 (tdata->td_winflags & TCP_WSCALE_SEEN)) { 1363 if (fr_tcpoptions(fin, tcp, fdata) == -1) 1364 fin->fin_flx |= FI_BAD; 1365 if (!(fdata->td_winflags & TCP_WSCALE_SEEN)) { 1366 fdata->td_winscale = 0; 1367 tdata->td_winscale = 0; 1368 } 1369 } 1370 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1371 fr_checknewisn(fin, is); 1372 } else if (flags == TH_SYN) { 1373 is->is_s0[source] = ntohl(tcp->th_seq) + 1; 1374 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 1375 if (fr_tcpoptions(fin, tcp, tdata) == -1) 1376 fin->fin_flx |= FI_BAD; 1377 } 1378 1379 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1380 fr_checknewisn(fin, is); 1381 1382 } 1383 ret = 1; 1384 } else 1385 fin->fin_flx |= FI_OOW; 1386 MUTEX_EXIT(&is->is_lock); 1387 return ret; 1388} 1389 1390 1391/* ------------------------------------------------------------------------ */ 1392/* Function: fr_checknewisn */ 1393/* Returns: Nil */ 1394/* Parameters: fin(I) - pointer to packet information */ 1395/* is(I) - pointer to master state structure */ 1396/* */ 1397/* Check to see if this TCP connection is expecting and needs a new */ 1398/* sequence number for a particular direction of the connection. */ 1399/* */ 1400/* NOTE: This does not actually change the sequence numbers, only gets new */ 1401/* one ready. */ 1402/* ------------------------------------------------------------------------ */ 1403static void fr_checknewisn(fin, is) 1404fr_info_t *fin; 1405ipstate_t *is; 1406{ 1407 u_32_t sumd, old, new; 1408 tcphdr_t *tcp; 1409 int i; 1410 1411 i = fin->fin_rev; 1412 tcp = fin->fin_dp; 1413 1414 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) || 1415 ((i == 1) && !(is->is_flags & IS_ISNACK))) { 1416 old = ntohl(tcp->th_seq); 1417 new = fr_newisn(fin); 1418 is->is_isninc[i] = new - old; 1419 CALC_SUMD(old, new, sumd); 1420 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16); 1421 1422 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK); 1423 } 1424} 1425 1426 1427/* ------------------------------------------------------------------------ */ 1428/* Function: fr_tcpinwindow */ 1429/* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */ 1430/* Parameters: fin(I) - pointer to packet information */ 1431/* fdata(I) - pointer to tcp state informatio (forward) */ 1432/* tdata(I) - pointer to tcp state informatio (reverse) */ 1433/* tcp(I) - pointer to TCP packet header */ 1434/* */ 1435/* Given a packet has matched addresses and ports, check to see if it is */ 1436/* within the TCP data window. In a show of generosity, allow packets that */ 1437/* are within the window space behind the current sequence # as well. */ 1438/* ------------------------------------------------------------------------ */ 1439int fr_tcpinwindow(fin, fdata, tdata, tcp, flags) 1440fr_info_t *fin; 1441tcpdata_t *fdata, *tdata; 1442tcphdr_t *tcp; 1443int flags; 1444{ 1445 tcp_seq seq, ack, end; 1446 int ackskew, tcpflags; 1447 u_32_t win, maxwin; 1448 int dsize, inseq; 1449 1450 /* 1451 * Find difference between last checked packet and this packet. 1452 */ 1453 tcpflags = tcp->th_flags; 1454 seq = ntohl(tcp->th_seq); 1455 ack = ntohl(tcp->th_ack); 1456 if (tcpflags & TH_SYN) 1457 win = ntohs(tcp->th_win); 1458 else 1459 win = ntohs(tcp->th_win) << fdata->td_winscale; 1460#if 0 1461 /* 1462 * XXX - This is a kludge is here because IPFilter doesn't track SACK 1463 * options in TCP packets. This is not a trivial to do if one is to 1464 * consider the performance impact of it. So instead, if the 1465 * receiver has said SACK is ok, double the allowed window size. 1466 * This is disabled for testing of another workaround for a problem 1467 * with Microsoft Windows - see below. 1468 */ 1469 if ((tdata->td_winflags & TCP_SACK_PERMIT) != 0) 1470 win *= 2; 1471#endif 1472 1473 /* 1474 * A window of 0 produces undesirable behaviour from this function. 1475 */ 1476 if (win == 0) 1477 win = 1; 1478 1479 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1480 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0); 1481 1482 /* 1483 * if window scaling is present, the scaling is only allowed 1484 * for windows not in the first SYN packet. In that packet the 1485 * window is 65535 to specify the largest window possible 1486 * for receivers not implementing the window scale option. 1487 * Currently, we do not assume TTCP here. That means that 1488 * if we see a second packet from a host (after the initial 1489 * SYN), we can assume that the receiver of the SYN did 1490 * already send back the SYN/ACK (and thus that we know if 1491 * the receiver also does window scaling) 1492 */ 1493 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) { 1494 if (tdata->td_winflags & TCP_WSCALE_SEEN) { 1495 fdata->td_winflags &= ~TCP_WSCALE_FIRST; 1496 fdata->td_maxwin = win; 1497 } else { 1498 fdata->td_winscale = 0; 1499 fdata->td_winflags &= ~(TCP_WSCALE_FIRST| 1500 TCP_WSCALE_SEEN); 1501 tdata->td_winscale = 0; 1502 tdata->td_winflags &= ~(TCP_WSCALE_FIRST| 1503 TCP_WSCALE_SEEN); 1504 } 1505 } 1506 1507 end = seq + dsize; 1508 1509 if ((fdata->td_end == 0) && 1510 (!(flags & IS_TCPFSM) || 1511 ((tcpflags & TH_OPENING) == TH_OPENING))) { 1512 /* 1513 * Must be a (outgoing) SYN-ACK in reply to a SYN. 1514 */ 1515 fdata->td_end = end - 1; 1516 fdata->td_maxwin = 1; 1517 fdata->td_maxend = end + win; 1518 } 1519 1520 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */ 1521 ack = tdata->td_end; 1522 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 1523 (ack == 0)) { 1524 /* gross hack to get around certain broken tcp stacks */ 1525 ack = tdata->td_end; 1526 } 1527 1528 maxwin = tdata->td_maxwin; 1529 ackskew = tdata->td_end - ack; 1530 1531 /* 1532 * Strict sequencing only allows in-order delivery. 1533 */ 1534 if ((flags & IS_STRICT) != 0) { 1535 if (seq != fdata->td_end) { 1536 return 0; 1537 } 1538 } 1539 1540#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 1541#define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 1542 inseq = 0; 1543 if ((SEQ_GE(fdata->td_maxend, end)) && 1544 (SEQ_GE(seq, fdata->td_end - maxwin)) && 1545/* XXX what about big packets */ 1546#define MAXACKWINDOW 66000 1547 (-ackskew <= (MAXACKWINDOW << fdata->td_winscale)) && 1548 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) { 1549 inseq = 1; 1550 /* 1551 * Microsoft Windows will send the next packet to the right of the 1552 * window if SACK is in use. 1553 */ 1554 } else if ((seq == fdata->td_maxend) && (ackskew == 0) && 1555 (fdata->td_winflags & TCP_SACK_PERMIT) && 1556 (tdata->td_winflags & TCP_SACK_PERMIT)) { 1557 inseq = 1; 1558 } 1559 1560 if (inseq) { 1561 /* if ackskew < 0 then this should be due to fragmented 1562 * packets. There is no way to know the length of the 1563 * total packet in advance. 1564 * We do know the total length from the fragment cache though. 1565 * Note however that there might be more sessions with 1566 * exactly the same source and destination parameters in the 1567 * state cache (and source and destination is the only stuff 1568 * that is saved in the fragment cache). Note further that 1569 * some TCP connections in the state cache are hashed with 1570 * sport and dport as well which makes it not worthwhile to 1571 * look for them. 1572 * Thus, when ackskew is negative but still seems to belong 1573 * to this session, we bump up the destinations end value. 1574 */ 1575 if (ackskew < 0) 1576 tdata->td_end = ack; 1577 1578 /* update max window seen */ 1579 if (fdata->td_maxwin < win) 1580 fdata->td_maxwin = win; 1581 if (SEQ_GT(end, fdata->td_end)) 1582 fdata->td_end = end; 1583 if (SEQ_GE(ack + win, tdata->td_maxend)) 1584 tdata->td_maxend = ack + win; 1585 return 1; 1586 } 1587 return 0; 1588} 1589 1590 1591/* ------------------------------------------------------------------------ */ 1592/* Function: fr_stclone */ 1593/* Returns: ipstate_t* - NULL == cloning failed, */ 1594/* else pointer to new state structure */ 1595/* Parameters: fin(I) - pointer to packet information */ 1596/* tcp(I) - pointer to TCP/UDP header */ 1597/* is(I) - pointer to master state structure */ 1598/* */ 1599/* Create a "duplcate" state table entry from the master. */ 1600/* ------------------------------------------------------------------------ */ 1601static ipstate_t *fr_stclone(fin, tcp, is) 1602fr_info_t *fin; 1603tcphdr_t *tcp; 1604ipstate_t *is; 1605{ 1606 ipstate_t *clone; 1607 u_32_t send; 1608 1609 if (ips_num == fr_statemax) { 1610 ATOMIC_INCL(ips_stats.iss_max); 1611 fr_state_doflush = 1; 1612 return NULL; 1613 } 1614 KMALLOC(clone, ipstate_t *); 1615 if (clone == NULL) 1616 return NULL; 1617 bcopy((char *)is, (char *)clone, sizeof(*clone)); 1618 1619 MUTEX_NUKE(&clone->is_lock); 1620 1621 clone->is_die = ONE_DAY + fr_ticks; 1622 clone->is_state[0] = 0; 1623 clone->is_state[1] = 0; 1624 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1625 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 1626 ((tcp->th_flags & TH_FIN) ? 1 : 0); 1627 1628 if (fin->fin_rev == 1) { 1629 clone->is_dend = send; 1630 clone->is_maxdend = send; 1631 clone->is_send = 0; 1632 clone->is_maxswin = 1; 1633 clone->is_maxdwin = ntohs(tcp->th_win); 1634 if (clone->is_maxdwin == 0) 1635 clone->is_maxdwin = 1; 1636 } else { 1637 clone->is_send = send; 1638 clone->is_maxsend = send; 1639 clone->is_dend = 0; 1640 clone->is_maxdwin = 1; 1641 clone->is_maxswin = ntohs(tcp->th_win); 1642 if (clone->is_maxswin == 0) 1643 clone->is_maxswin = 1; 1644 } 1645 1646 clone->is_flags &= ~SI_CLONE; 1647 clone->is_flags |= SI_CLONED; 1648 fr_stinsert(clone, fin->fin_rev); 1649 clone->is_ref = 2; 1650 if (clone->is_p == IPPROTO_TCP) { 1651 (void) fr_tcp_age(&clone->is_sti, fin, ips_tqtqb, 1652 clone->is_flags); 1653 } 1654 MUTEX_EXIT(&clone->is_lock); 1655#ifdef IPFILTER_SCAN 1656 (void) ipsc_attachis(is); 1657#endif 1658#ifdef IPFILTER_SYNC 1659 if (is->is_flags & IS_STATESYNC) 1660 clone->is_sync = ipfsync_new(SMC_STATE, fin, clone); 1661#endif 1662 return clone; 1663} 1664 1665 1666/* ------------------------------------------------------------------------ */ 1667/* Function: fr_matchsrcdst */ 1668/* Returns: Nil */ 1669/* Parameters: fin(I) - pointer to packet information */ 1670/* is(I) - pointer to state structure */ 1671/* src(I) - pointer to source address */ 1672/* dst(I) - pointer to destination address */ 1673/* tcp(I) - pointer to TCP/UDP header */ 1674/* */ 1675/* Match a state table entry against an IP packet. The logic below is that */ 1676/* ret gets set to one if the match succeeds, else remains 0. If it is */ 1677/* still 0 after the test. no match. */ 1678/* ------------------------------------------------------------------------ */ 1679static ipstate_t *fr_matchsrcdst(fin, is, src, dst, tcp, cmask) 1680fr_info_t *fin; 1681ipstate_t *is; 1682i6addr_t *src, *dst; 1683tcphdr_t *tcp; 1684u_32_t cmask; 1685{ 1686 int ret = 0, rev, out, flags, flx = 0, idx; 1687 u_short sp, dp; 1688 u_32_t cflx; 1689 void *ifp; 1690 1691 rev = IP6_NEQ(&is->is_dst, dst); 1692 ifp = fin->fin_ifp; 1693 out = fin->fin_out; 1694 flags = is->is_flags; 1695 sp = 0; 1696 dp = 0; 1697 1698 if (tcp != NULL) { 1699 sp = htons(fin->fin_sport); 1700 dp = ntohs(fin->fin_dport); 1701 } 1702 if (!rev) { 1703 if (tcp != NULL) { 1704 if (!(flags & SI_W_SPORT) && (sp != is->is_sport)) 1705 rev = 1; 1706 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport)) 1707 rev = 1; 1708 } 1709 } 1710 1711 idx = (out << 1) + rev; 1712 1713 /* 1714 * If the interface for this 'direction' is set, make sure it matches. 1715 * An interface name that is not set matches any, as does a name of *. 1716 */ 1717 if ((is->is_ifp[idx] == NULL && 1718 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) || 1719 is->is_ifp[idx] == ifp) 1720 ret = 1; 1721 1722 if (ret == 0) 1723 return NULL; 1724 ret = 0; 1725 1726 /* 1727 * Match addresses and ports. 1728 */ 1729 if (rev == 0) { 1730 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) && 1731 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) { 1732 if (tcp) { 1733 if ((sp == is->is_sport || flags & SI_W_SPORT)&& 1734 (dp == is->is_dport || flags & SI_W_DPORT)) 1735 ret = 1; 1736 } else { 1737 ret = 1; 1738 } 1739 } 1740 } else { 1741 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) && 1742 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) { 1743 if (tcp) { 1744 if ((dp == is->is_sport || flags & SI_W_SPORT)&& 1745 (sp == is->is_dport || flags & SI_W_DPORT)) 1746 ret = 1; 1747 } else { 1748 ret = 1; 1749 } 1750 } 1751 } 1752 1753 if (ret == 0) 1754 return NULL; 1755 1756 /* 1757 * Whether or not this should be here, is questionable, but the aim 1758 * is to get this out of the main line. 1759 */ 1760 if (tcp == NULL) 1761 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED); 1762 1763 /* 1764 * Only one of the source or destination address can be flaged as a 1765 * wildcard. Fill in the missing address, if set. 1766 * For IPv6, if the address being copied in is multicast, then 1767 * don't reset the wild flag - multicast causes it to be set in the 1768 * first place! 1769 */ 1770 if ((flags & (SI_W_SADDR|SI_W_DADDR))) { 1771 fr_ip_t *fi = &fin->fin_fi; 1772 1773 if ((flags & SI_W_SADDR) != 0) { 1774 if (rev == 0) { 1775#ifdef USE_INET6 1776 if (is->is_v == 6 && 1777 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 1778 /*EMPTY*/; 1779 else 1780#endif 1781 { 1782 is->is_src = fi->fi_src; 1783 is->is_flags &= ~SI_W_SADDR; 1784 } 1785 } else { 1786#ifdef USE_INET6 1787 if (is->is_v == 6 && 1788 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 1789 /*EMPTY*/; 1790 else 1791#endif 1792 { 1793 is->is_src = fi->fi_dst; 1794 is->is_flags &= ~SI_W_SADDR; 1795 } 1796 } 1797 } else if ((flags & SI_W_DADDR) != 0) { 1798 if (rev == 0) { 1799#ifdef USE_INET6 1800 if (is->is_v == 6 && 1801 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 1802 /*EMPTY*/; 1803 else 1804#endif 1805 { 1806 is->is_dst = fi->fi_dst; 1807 is->is_flags &= ~SI_W_DADDR; 1808 } 1809 } else { 1810#ifdef USE_INET6 1811 if (is->is_v == 6 && 1812 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 1813 /*EMPTY*/; 1814 else 1815#endif 1816 { 1817 is->is_dst = fi->fi_src; 1818 is->is_flags &= ~SI_W_DADDR; 1819 } 1820 } 1821 } 1822 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) { 1823 ATOMIC_DECL(ips_stats.iss_wild); 1824 } 1825 } 1826 1827 flx = fin->fin_flx & cmask; 1828 cflx = is->is_flx[out][rev]; 1829 1830 /* 1831 * Match up any flags set from IP options. 1832 */ 1833 if ((cflx && (flx != (cflx & cmask))) || 1834 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) || 1835 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) || 1836 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) 1837 return NULL; 1838 1839 /* 1840 * Only one of the source or destination port can be flagged as a 1841 * wildcard. When filling it in, fill in a copy of the matched entry 1842 * if it has the cloning flag set. 1843 */ 1844 if ((fin->fin_flx & FI_IGNORE) != 0) { 1845 fin->fin_rev = rev; 1846 return is; 1847 } 1848 1849 if ((flags & (SI_W_SPORT|SI_W_DPORT))) { 1850 if ((flags & SI_CLONE) != 0) { 1851 ipstate_t *clone; 1852 1853 clone = fr_stclone(fin, tcp, is); 1854 if (clone == NULL) 1855 return NULL; 1856 is = clone; 1857 } else { 1858 ATOMIC_DECL(ips_stats.iss_wild); 1859 } 1860 1861 if ((flags & SI_W_SPORT) != 0) { 1862 if (rev == 0) { 1863 is->is_sport = sp; 1864 is->is_send = ntohl(tcp->th_seq); 1865 } else { 1866 is->is_sport = dp; 1867 is->is_send = ntohl(tcp->th_ack); 1868 } 1869 is->is_maxsend = is->is_send + 1; 1870 } else if ((flags & SI_W_DPORT) != 0) { 1871 if (rev == 0) { 1872 is->is_dport = dp; 1873 is->is_dend = ntohl(tcp->th_ack); 1874 } else { 1875 is->is_dport = sp; 1876 is->is_dend = ntohl(tcp->th_seq); 1877 } 1878 is->is_maxdend = is->is_dend + 1; 1879 } 1880 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT); 1881 if ((flags & SI_CLONED) && ipstate_logging) 1882 ipstate_log(is, ISL_CLONE); 1883 } 1884 1885 ret = -1; 1886 1887 if (is->is_flx[out][rev] == 0) { 1888 is->is_flx[out][rev] = flx; 1889 is->is_opt[rev] = fin->fin_optmsk; 1890 if (is->is_v == 6) { 1891 is->is_opt[rev] &= ~0x8; 1892 is->is_optmsk[rev] &= ~0x8; 1893 } 1894 } 1895 1896 /* 1897 * Check if the interface name for this "direction" is set and if not, 1898 * fill it in. 1899 */ 1900 if (is->is_ifp[idx] == NULL && 1901 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) { 1902 is->is_ifp[idx] = ifp; 1903 COPYIFNAME(ifp, is->is_ifname[idx]); 1904 } 1905 fin->fin_rev = rev; 1906 return is; 1907} 1908 1909 1910/* ------------------------------------------------------------------------ */ 1911/* Function: fr_checkicmpmatchingstate */ 1912/* Returns: Nil */ 1913/* Parameters: fin(I) - pointer to packet information */ 1914/* */ 1915/* If we've got an ICMP error message, using the information stored in the */ 1916/* ICMP packet, look for a matching state table entry. */ 1917/* */ 1918/* If we return NULL then no lock on ipf_state is held. */ 1919/* If we return non-null then a read-lock on ipf_state is held. */ 1920/* ------------------------------------------------------------------------ */ 1921static ipstate_t *fr_checkicmpmatchingstate(fin) 1922fr_info_t *fin; 1923{ 1924 ipstate_t *is, **isp; 1925 u_short sport, dport; 1926 u_char pr; 1927 int backward, i, oi; 1928 i6addr_t dst, src; 1929 struct icmp *ic; 1930 u_short savelen; 1931 icmphdr_t *icmp; 1932 fr_info_t ofin; 1933 tcphdr_t *tcp; 1934 int type, len; 1935 ip_t *oip; 1936 u_int hv; 1937 1938 /* 1939 * Does it at least have the return (basic) IP header ? 1940 * Is it an actual recognised ICMP error type? 1941 * Only a basic IP header (no options) should be with 1942 * an ICMP error header. 1943 */ 1944 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) || 1945 (fin->fin_plen < ICMPERR_MINPKTLEN) || 1946 !(fin->fin_flx & FI_ICMPERR)) 1947 return NULL; 1948 ic = fin->fin_dp; 1949 type = ic->icmp_type; 1950 1951 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 1952 /* 1953 * Check if the at least the old IP header (with options) and 1954 * 8 bytes of payload is present. 1955 */ 1956 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) 1957 return NULL; 1958 1959 /* 1960 * Sanity Checks. 1961 */ 1962 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 1963 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) 1964 return NULL; 1965 1966 /* 1967 * Is the buffer big enough for all of it ? It's the size of the IP 1968 * header claimed in the encapsulated part which is of concern. It 1969 * may be too big to be in this buffer but not so big that it's 1970 * outside the ICMP packet, leading to TCP deref's causing problems. 1971 * This is possible because we don't know how big oip_hl is when we 1972 * do the pullup early in fr_check() and thus can't guarantee it is 1973 * all here now. 1974 */ 1975#ifdef _KERNEL 1976 { 1977 mb_t *m; 1978 1979 m = fin->fin_m; 1980# if defined(MENTAT) 1981 if ((char *)oip + len > (char *)m->b_wptr) 1982 return NULL; 1983# else 1984 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) 1985 return NULL; 1986# endif 1987 } 1988#endif 1989 bcopy((char *)fin, (char *)&ofin, sizeof(fin)); 1990 1991 /* 1992 * in the IPv4 case we must zero the i6addr union otherwise 1993 * the IP6_EQ and IP6_NEQ macros produce the wrong results because 1994 * of the 'junk' in the unused part of the union 1995 */ 1996 bzero((char *)&src, sizeof(src)); 1997 bzero((char *)&dst, sizeof(dst)); 1998 1999 /* 2000 * we make an fin entry to be able to feed it to 2001 * matchsrcdst note that not all fields are encessary 2002 * but this is the cleanest way. Note further we fill 2003 * in fin_mp such that if someone uses it we'll get 2004 * a kernel panic. fr_matchsrcdst does not use this. 2005 * 2006 * watch out here, as ip is in host order and oip in network 2007 * order. Any change we make must be undone afterwards, like 2008 * oip->ip_off - it is still in network byte order so fix it. 2009 */ 2010 savelen = oip->ip_len; 2011 oip->ip_len = len; 2012 oip->ip_off = ntohs(oip->ip_off); 2013 2014 ofin.fin_flx = FI_NOCKSUM; 2015 ofin.fin_v = 4; 2016 ofin.fin_ip = oip; 2017 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 2018 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 2019 ofin.fin_plen = fin->fin_dlen - ICMPERR_ICMPHLEN; 2020 (void) fr_makefrip(IP_HL(oip) << 2, oip, &ofin); 2021 ofin.fin_ifp = fin->fin_ifp; 2022 ofin.fin_out = !fin->fin_out; 2023 /* 2024 * Reset the short and bad flag here because in fr_matchsrcdst() 2025 * the flags for the current packet (fin_flx) are compared against 2026 * those for the existing session. 2027 */ 2028 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 2029 2030 /* 2031 * Put old values of ip_len and ip_off back as we don't know 2032 * if we have to forward the packet (or process it again. 2033 */ 2034 oip->ip_len = savelen; 2035 oip->ip_off = htons(oip->ip_off); 2036 2037 switch (oip->ip_p) 2038 { 2039 case IPPROTO_ICMP : 2040 /* 2041 * an ICMP error can only be generated as a result of an 2042 * ICMP query, not as the response on an ICMP error 2043 * 2044 * XXX theoretically ICMP_ECHOREP and the other reply's are 2045 * ICMP query's as well, but adding them here seems strange XXX 2046 */ 2047 if ((ofin.fin_flx & FI_ICMPERR) != 0) 2048 return NULL; 2049 2050 /* 2051 * perform a lookup of the ICMP packet in the state table 2052 */ 2053 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2054 hv = (pr = oip->ip_p); 2055 src.in4 = oip->ip_src; 2056 hv += src.in4.s_addr; 2057 dst.in4 = oip->ip_dst; 2058 hv += dst.in4.s_addr; 2059 hv += icmp->icmp_id; 2060 hv = DOUBLE_HASH(hv); 2061 2062 READ_ENTER(&ipf_state); 2063 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2064 isp = &is->is_hnext; 2065 if ((is->is_p != pr) || (is->is_v != 4)) 2066 continue; 2067 if (is->is_pass & FR_NOICMPERR) 2068 continue; 2069 is = fr_matchsrcdst(&ofin, is, &src, &dst, 2070 NULL, FI_ICMPCMP); 2071 if (is != NULL) { 2072 /* 2073 * i : the index of this packet (the icmp 2074 * unreachable) 2075 * oi : the index of the original packet found 2076 * in the icmp header (i.e. the packet 2077 * causing this icmp) 2078 * backward : original packet was backward 2079 * compared to the state 2080 */ 2081 backward = IP6_NEQ(&is->is_src, &src); 2082 fin->fin_rev = !backward; 2083 i = (!backward << 1) + fin->fin_out; 2084 oi = (backward << 1) + ofin.fin_out; 2085 if (is->is_icmppkts[i] > is->is_pkts[oi]) 2086 continue; 2087 ips_stats.iss_hits++; 2088 is->is_icmppkts[i]++; 2089 return is; 2090 } 2091 } 2092 RWLOCK_EXIT(&ipf_state); 2093 return NULL; 2094 case IPPROTO_TCP : 2095 case IPPROTO_UDP : 2096 break; 2097 default : 2098 return NULL; 2099 } 2100 2101 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2102 dport = tcp->th_dport; 2103 sport = tcp->th_sport; 2104 2105 hv = (pr = oip->ip_p); 2106 src.in4 = oip->ip_src; 2107 hv += src.in4.s_addr; 2108 dst.in4 = oip->ip_dst; 2109 hv += dst.in4.s_addr; 2110 hv += dport; 2111 hv += sport; 2112 hv = DOUBLE_HASH(hv); 2113 2114 READ_ENTER(&ipf_state); 2115 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2116 isp = &is->is_hnext; 2117 /* 2118 * Only allow this icmp though if the 2119 * encapsulated packet was allowed through the 2120 * other way around. Note that the minimal amount 2121 * of info present does not allow for checking against 2122 * tcp internals such as seq and ack numbers. Only the 2123 * ports are known to be present and can be even if the 2124 * short flag is set. 2125 */ 2126 if ((is->is_p == pr) && (is->is_v == 4) && 2127 (is = fr_matchsrcdst(&ofin, is, &src, &dst, 2128 tcp, FI_ICMPCMP))) { 2129 /* 2130 * i : the index of this packet (the icmp unreachable) 2131 * oi : the index of the original packet found in the 2132 * icmp header (i.e. the packet causing this icmp) 2133 * backward : original packet was backward compared to 2134 * the state 2135 */ 2136 backward = IP6_NEQ(&is->is_src, &src); 2137 fin->fin_rev = !backward; 2138 i = (!backward << 1) + fin->fin_out; 2139 oi = (backward << 1) + ofin.fin_out; 2140 2141 if (((is->is_pass & FR_NOICMPERR) != 0) || 2142 (is->is_icmppkts[i] > is->is_pkts[oi])) 2143 break; 2144 ips_stats.iss_hits++; 2145 is->is_icmppkts[i]++; 2146 /* 2147 * we deliberately do not touch the timeouts 2148 * for the accompanying state table entry. 2149 * It remains to be seen if that is correct. XXX 2150 */ 2151 return is; 2152 } 2153 } 2154 RWLOCK_EXIT(&ipf_state); 2155 return NULL; 2156} 2157 2158 2159/* ------------------------------------------------------------------------ */ 2160/* Function: fr_ipsmove */ 2161/* Returns: Nil */ 2162/* Parameters: is(I) - pointer to state table entry */ 2163/* hv(I) - new hash value for state table entry */ 2164/* Write Locks: ipf_state */ 2165/* */ 2166/* Move a state entry from one position in the hash table to another. */ 2167/* ------------------------------------------------------------------------ */ 2168static void fr_ipsmove(is, hv) 2169ipstate_t *is; 2170u_int hv; 2171{ 2172 ipstate_t **isp; 2173 u_int hvm; 2174 2175 ASSERT(rw_read_locked(&ipf_state.ipf_lk) == 0); 2176 2177 hvm = is->is_hv; 2178 /* 2179 * Remove the hash from the old location... 2180 */ 2181 isp = is->is_phnext; 2182 if (is->is_hnext) 2183 is->is_hnext->is_phnext = isp; 2184 *isp = is->is_hnext; 2185 if (ips_table[hvm] == NULL) 2186 ips_stats.iss_inuse--; 2187 ips_stats.iss_bucketlen[hvm]--; 2188 2189 /* 2190 * ...and put the hash in the new one. 2191 */ 2192 hvm = DOUBLE_HASH(hv); 2193 is->is_hv = hvm; 2194 isp = &ips_table[hvm]; 2195 if (*isp) 2196 (*isp)->is_phnext = &is->is_hnext; 2197 else 2198 ips_stats.iss_inuse++; 2199 ips_stats.iss_bucketlen[hvm]++; 2200 is->is_phnext = isp; 2201 is->is_hnext = *isp; 2202 *isp = is; 2203} 2204 2205 2206/* ------------------------------------------------------------------------ */ 2207/* Function: fr_stlookup */ 2208/* Returns: ipstate_t* - NULL == no matching state found, */ 2209/* else pointer to state information is returned */ 2210/* Parameters: fin(I) - pointer to packet information */ 2211/* tcp(I) - pointer to TCP/UDP header. */ 2212/* */ 2213/* Search the state table for a matching entry to the packet described by */ 2214/* the contents of *fin. */ 2215/* */ 2216/* If we return NULL then no lock on ipf_state is held. */ 2217/* If we return non-null then a read-lock on ipf_state is held. */ 2218/* ------------------------------------------------------------------------ */ 2219ipstate_t *fr_stlookup(fin, tcp, ifqp) 2220fr_info_t *fin; 2221tcphdr_t *tcp; 2222ipftq_t **ifqp; 2223{ 2224 u_int hv, hvm, pr, v, tryagain; 2225 ipstate_t *is, **isp; 2226 u_short dport, sport; 2227 i6addr_t src, dst; 2228 struct icmp *ic; 2229 ipftq_t *ifq; 2230 int oow; 2231 2232 is = NULL; 2233 ifq = NULL; 2234 tcp = fin->fin_dp; 2235 ic = (struct icmp *)tcp; 2236 hv = (pr = fin->fin_fi.fi_p); 2237 src = fin->fin_fi.fi_src; 2238 dst = fin->fin_fi.fi_dst; 2239 hv += src.in4.s_addr; 2240 hv += dst.in4.s_addr; 2241 2242 v = fin->fin_fi.fi_v; 2243#ifdef USE_INET6 2244 if (v == 6) { 2245 hv += fin->fin_fi.fi_src.i6[1]; 2246 hv += fin->fin_fi.fi_src.i6[2]; 2247 hv += fin->fin_fi.fi_src.i6[3]; 2248 2249 if ((fin->fin_p == IPPROTO_ICMPV6) && 2250 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) { 2251 hv -= dst.in4.s_addr; 2252 } else { 2253 hv += fin->fin_fi.fi_dst.i6[1]; 2254 hv += fin->fin_fi.fi_dst.i6[2]; 2255 hv += fin->fin_fi.fi_dst.i6[3]; 2256 } 2257 } 2258#endif 2259 2260 /* 2261 * Search the hash table for matching packet header info. 2262 */ 2263 switch (pr) 2264 { 2265#ifdef USE_INET6 2266 case IPPROTO_ICMPV6 : 2267 tryagain = 0; 2268 if (v == 6) { 2269 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 2270 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 2271 hv += ic->icmp_id; 2272 } 2273 } 2274 READ_ENTER(&ipf_state); 2275icmp6again: 2276 hvm = DOUBLE_HASH(hv); 2277 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2278 isp = &is->is_hnext; 2279 if ((is->is_p != pr) || (is->is_v != v)) 2280 continue; 2281 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2282 if (is != NULL && 2283 fr_matchicmpqueryreply(v, &is->is_icmp, 2284 ic, fin->fin_rev)) { 2285 if (fin->fin_rev) 2286 ifq = &ips_icmpacktq; 2287 else 2288 ifq = &ips_icmptq; 2289 break; 2290 } 2291 } 2292 2293 if (is != NULL) { 2294 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) { 2295 hv += fin->fin_fi.fi_src.i6[0]; 2296 hv += fin->fin_fi.fi_src.i6[1]; 2297 hv += fin->fin_fi.fi_src.i6[2]; 2298 hv += fin->fin_fi.fi_src.i6[3]; 2299 fr_ipsmove(is, hv); 2300 MUTEX_DOWNGRADE(&ipf_state); 2301 } 2302 break; 2303 } 2304 RWLOCK_EXIT(&ipf_state); 2305 2306 /* 2307 * No matching icmp state entry. Perhaps this is a 2308 * response to another state entry. 2309 * 2310 * XXX With some ICMP6 packets, the "other" address is already 2311 * in the packet, after the ICMP6 header, and this could be 2312 * used in place of the multicast address. However, taking 2313 * advantage of this requires some significant code changes 2314 * to handle the specific types where that is the case. 2315 */ 2316 if ((ips_stats.iss_wild != 0) && (v == 6) && (tryagain == 0) && 2317 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) { 2318 hv -= fin->fin_fi.fi_src.i6[0]; 2319 hv -= fin->fin_fi.fi_src.i6[1]; 2320 hv -= fin->fin_fi.fi_src.i6[2]; 2321 hv -= fin->fin_fi.fi_src.i6[3]; 2322 tryagain = 1; 2323 WRITE_ENTER(&ipf_state); 2324 goto icmp6again; 2325 } 2326 2327 is = fr_checkicmp6matchingstate(fin); 2328 if (is != NULL) 2329 return is; 2330 break; 2331#endif 2332 2333 case IPPROTO_ICMP : 2334 if (v == 4) { 2335 hv += ic->icmp_id; 2336 } 2337 hv = DOUBLE_HASH(hv); 2338 READ_ENTER(&ipf_state); 2339 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2340 isp = &is->is_hnext; 2341 if ((is->is_p != pr) || (is->is_v != v)) 2342 continue; 2343 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2344 if (is != NULL && 2345 fr_matchicmpqueryreply(v, &is->is_icmp, 2346 ic, fin->fin_rev)) { 2347 if (fin->fin_rev) 2348 ifq = &ips_icmpacktq; 2349 else 2350 ifq = &ips_icmptq; 2351 break; 2352 } 2353 } 2354 if (is == NULL) { 2355 RWLOCK_EXIT(&ipf_state); 2356 } 2357 break; 2358 2359 case IPPROTO_TCP : 2360 case IPPROTO_UDP : 2361 ifqp = NULL; 2362 sport = htons(fin->fin_data[0]); 2363 hv += sport; 2364 dport = htons(fin->fin_data[1]); 2365 hv += dport; 2366 oow = 0; 2367 tryagain = 0; 2368 READ_ENTER(&ipf_state); 2369retry_tcpudp: 2370 hvm = DOUBLE_HASH(hv); 2371 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2372 isp = &is->is_hnext; 2373 if ((is->is_p != pr) || (is->is_v != v)) 2374 continue; 2375 fin->fin_flx &= ~FI_OOW; 2376 is = fr_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP); 2377 if (is != NULL) { 2378 if (pr == IPPROTO_TCP) { 2379 if (!fr_tcpstate(fin, tcp, is)) { 2380 oow |= fin->fin_flx & FI_OOW; 2381 continue; 2382 } 2383 } 2384 break; 2385 } 2386 } 2387 if (is != NULL) { 2388 if (tryagain && 2389 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) { 2390 hv += dport; 2391 hv += sport; 2392 fr_ipsmove(is, hv); 2393 MUTEX_DOWNGRADE(&ipf_state); 2394 } 2395 break; 2396 } 2397 RWLOCK_EXIT(&ipf_state); 2398 2399 if (!tryagain && ips_stats.iss_wild) { 2400 hv -= dport; 2401 hv -= sport; 2402 tryagain = 1; 2403 WRITE_ENTER(&ipf_state); 2404 goto retry_tcpudp; 2405 } 2406 fin->fin_flx |= oow; 2407 break; 2408 2409#if 0 2410 case IPPROTO_GRE : 2411 gre = fin->fin_dp; 2412 if (GRE_REV(gre->gr_flags) == 1) { 2413 hv += gre->gr_call; 2414 } 2415 /* FALLTHROUGH */ 2416#endif 2417 default : 2418 ifqp = NULL; 2419 hvm = DOUBLE_HASH(hv); 2420 READ_ENTER(&ipf_state); 2421 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2422 isp = &is->is_hnext; 2423 if ((is->is_p != pr) || (is->is_v != v)) 2424 continue; 2425 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2426 if (is != NULL) { 2427 ifq = &ips_iptq; 2428 break; 2429 } 2430 } 2431 if (is == NULL) { 2432 RWLOCK_EXIT(&ipf_state); 2433 } 2434 break; 2435 } 2436 2437 if ((is != NULL) && ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) && 2438 (is->is_tqehead[fin->fin_rev] != NULL)) 2439 ifq = is->is_tqehead[fin->fin_rev]; 2440 if (ifq != NULL && ifqp != NULL) 2441 *ifqp = ifq; 2442 return is; 2443} 2444 2445 2446/* ------------------------------------------------------------------------ */ 2447/* Function: fr_updatestate */ 2448/* Returns: Nil */ 2449/* Parameters: fin(I) - pointer to packet information */ 2450/* is(I) - pointer to state table entry */ 2451/* Read Locks: ipf_state */ 2452/* */ 2453/* Updates packet and byte counters for a newly received packet. Seeds the */ 2454/* fragment cache with a new entry as required. */ 2455/* ------------------------------------------------------------------------ */ 2456void fr_updatestate(fin, is, ifq) 2457fr_info_t *fin; 2458ipstate_t *is; 2459ipftq_t *ifq; 2460{ 2461 ipftqent_t *tqe; 2462 int i, pass; 2463 2464 i = (fin->fin_rev << 1) + fin->fin_out; 2465 2466 /* 2467 * For TCP packets, ifq == NULL. For all others, check if this new 2468 * queue is different to the last one it was on and move it if so. 2469 */ 2470 tqe = &is->is_sti; 2471 MUTEX_ENTER(&is->is_lock); 2472 if ((tqe->tqe_flags & TQE_RULEBASED) != 0) 2473 ifq = is->is_tqehead[fin->fin_rev]; 2474 2475 if (ifq != NULL) 2476 fr_movequeue(tqe, tqe->tqe_ifq, ifq); 2477 2478 is->is_pkts[i]++; 2479 is->is_bytes[i] += fin->fin_plen; 2480 MUTEX_EXIT(&is->is_lock); 2481 2482#ifdef IPFILTER_SYNC 2483 if (is->is_flags & IS_STATESYNC) 2484 ipfsync_update(SMC_STATE, fin, is->is_sync); 2485#endif 2486 2487 ATOMIC_INCL(ips_stats.iss_hits); 2488 2489 fin->fin_fr = is->is_rule; 2490 2491 /* 2492 * If this packet is a fragment and the rule says to track fragments, 2493 * then create a new fragment cache entry. 2494 */ 2495 pass = is->is_pass; 2496 if ((fin->fin_flx & FI_FRAG) && FR_ISPASS(pass)) 2497 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 2498} 2499 2500 2501/* ------------------------------------------------------------------------ */ 2502/* Function: fr_checkstate */ 2503/* Returns: frentry_t* - NULL == search failed, */ 2504/* else pointer to rule for matching state */ 2505/* Parameters: ifp(I) - pointer to interface */ 2506/* passp(I) - pointer to filtering result flags */ 2507/* */ 2508/* Check if a packet is associated with an entry in the state table. */ 2509/* ------------------------------------------------------------------------ */ 2510frentry_t *fr_checkstate(fin, passp) 2511fr_info_t *fin; 2512u_32_t *passp; 2513{ 2514 ipstate_t *is; 2515 frentry_t *fr; 2516 tcphdr_t *tcp; 2517 ipftq_t *ifq; 2518 u_int pass; 2519 2520 if (fr_state_lock || (ips_list == NULL) || 2521 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 2522 return NULL; 2523 2524 is = NULL; 2525 if ((fin->fin_flx & FI_TCPUDP) || 2526 (fin->fin_fi.fi_p == IPPROTO_ICMP) 2527#ifdef USE_INET6 2528 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6) 2529#endif 2530 ) 2531 tcp = fin->fin_dp; 2532 else 2533 tcp = NULL; 2534 2535 /* 2536 * Search the hash table for matching packet header info. 2537 */ 2538 ifq = NULL; 2539 is = fin->fin_state; 2540 if (is == NULL) 2541 is = fr_stlookup(fin, tcp, &ifq); 2542 switch (fin->fin_p) 2543 { 2544#ifdef USE_INET6 2545 case IPPROTO_ICMPV6 : 2546 if (is != NULL) 2547 break; 2548 if (fin->fin_v == 6) { 2549 is = fr_checkicmp6matchingstate(fin); 2550 if (is != NULL) 2551 goto matched; 2552 } 2553 break; 2554#endif 2555 case IPPROTO_ICMP : 2556 if (is != NULL) 2557 break; 2558 /* 2559 * No matching icmp state entry. Perhaps this is a 2560 * response to another state entry. 2561 */ 2562 is = fr_checkicmpmatchingstate(fin); 2563 if (is != NULL) 2564 goto matched; 2565 break; 2566 case IPPROTO_TCP : 2567 if (is == NULL) 2568 break; 2569 2570 if (is->is_pass & FR_NEWISN) { 2571 if (fin->fin_out == 0) 2572 fr_fixinisn(fin, is); 2573 else if (fin->fin_out == 1) 2574 fr_fixoutisn(fin, is); 2575 } 2576 break; 2577 default : 2578 if (fin->fin_rev) 2579 ifq = &ips_udpacktq; 2580 else 2581 ifq = &ips_udptq; 2582 break; 2583 } 2584 if (is == NULL) { 2585 ATOMIC_INCL(ips_stats.iss_miss); 2586 return NULL; 2587 } 2588 2589matched: 2590 fr = is->is_rule; 2591 if (fr != NULL) { 2592 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) { 2593 if (fin->fin_nattag == NULL) 2594 return NULL; 2595 if (fr_matchtag(&fr->fr_nattag, fin->fin_nattag) != 0) 2596 return NULL; 2597 } 2598 (void) strncpy(fin->fin_group, fr->fr_group, FR_GROUPLEN); 2599 fin->fin_icode = fr->fr_icode; 2600 } 2601 2602 fin->fin_rule = is->is_rulen; 2603 pass = is->is_pass; 2604 fr_updatestate(fin, is, ifq); 2605 if (fin->fin_out == 1) 2606 fin->fin_nat = is->is_nat[fin->fin_rev]; 2607 2608 fin->fin_state = is; 2609 is->is_touched = fr_ticks; 2610 MUTEX_ENTER(&is->is_lock); 2611 is->is_ref++; 2612 MUTEX_EXIT(&is->is_lock); 2613 RWLOCK_EXIT(&ipf_state); 2614 fin->fin_flx |= FI_STATE; 2615 if ((pass & FR_LOGFIRST) != 0) 2616 pass &= ~(FR_LOGFIRST|FR_LOG); 2617 *passp = pass; 2618 return fr; 2619} 2620 2621 2622/* ------------------------------------------------------------------------ */ 2623/* Function: fr_fixoutisn */ 2624/* Returns: Nil */ 2625/* Parameters: fin(I) - pointer to packet information */ 2626/* is(I) - pointer to master state structure */ 2627/* */ 2628/* Called only for outbound packets, adjusts the sequence number and the */ 2629/* TCP checksum to match that change. */ 2630/* ------------------------------------------------------------------------ */ 2631static void fr_fixoutisn(fin, is) 2632fr_info_t *fin; 2633ipstate_t *is; 2634{ 2635 tcphdr_t *tcp; 2636 int rev; 2637 u_32_t seq; 2638 2639 tcp = fin->fin_dp; 2640 rev = fin->fin_rev; 2641 if ((is->is_flags & IS_ISNSYN) != 0) { 2642 if (rev == 0) { 2643 seq = ntohl(tcp->th_seq); 2644 seq += is->is_isninc[0]; 2645 tcp->th_seq = htonl(seq); 2646 fix_outcksum(fin, &tcp->th_sum, is->is_sumd[0]); 2647 } 2648 } 2649 if ((is->is_flags & IS_ISNACK) != 0) { 2650 if (rev == 1) { 2651 seq = ntohl(tcp->th_seq); 2652 seq += is->is_isninc[1]; 2653 tcp->th_seq = htonl(seq); 2654 fix_outcksum(fin, &tcp->th_sum, is->is_sumd[1]); 2655 } 2656 } 2657} 2658 2659 2660/* ------------------------------------------------------------------------ */ 2661/* Function: fr_fixinisn */ 2662/* Returns: Nil */ 2663/* Parameters: fin(I) - pointer to packet information */ 2664/* is(I) - pointer to master state structure */ 2665/* */ 2666/* Called only for inbound packets, adjusts the acknowledge number and the */ 2667/* TCP checksum to match that change. */ 2668/* ------------------------------------------------------------------------ */ 2669static void fr_fixinisn(fin, is) 2670fr_info_t *fin; 2671ipstate_t *is; 2672{ 2673 tcphdr_t *tcp; 2674 int rev; 2675 u_32_t ack; 2676 2677 tcp = fin->fin_dp; 2678 rev = fin->fin_rev; 2679 if ((is->is_flags & IS_ISNSYN) != 0) { 2680 if (rev == 1) { 2681 ack = ntohl(tcp->th_ack); 2682 ack -= is->is_isninc[0]; 2683 tcp->th_ack = htonl(ack); 2684 fix_incksum(fin, &tcp->th_sum, is->is_sumd[0]); 2685 } 2686 } 2687 if ((is->is_flags & IS_ISNACK) != 0) { 2688 if (rev == 0) { 2689 ack = ntohl(tcp->th_ack); 2690 ack -= is->is_isninc[1]; 2691 tcp->th_ack = htonl(ack); 2692 fix_incksum(fin, &tcp->th_sum, is->is_sumd[1]); 2693 } 2694 } 2695} 2696 2697 2698/* ------------------------------------------------------------------------ */ 2699/* Function: fr_statesync */ 2700/* Returns: Nil */ 2701/* Parameters: ifp(I) - pointer to interface */ 2702/* */ 2703/* Walk through all state entries and if an interface pointer match is */ 2704/* found then look it up again, based on its name in case the pointer has */ 2705/* changed since last time. */ 2706/* */ 2707/* If ifp is passed in as being non-null then we are only doing updates for */ 2708/* existing, matching, uses of it. */ 2709/* ------------------------------------------------------------------------ */ 2710void fr_statesync(ifp) 2711void *ifp; 2712{ 2713 ipstate_t *is; 2714 int i; 2715 2716 if (fr_running <= 0) 2717 return; 2718 2719 WRITE_ENTER(&ipf_state); 2720 2721 if (fr_running <= 0) { 2722 RWLOCK_EXIT(&ipf_state); 2723 return; 2724 } 2725 2726 for (is = ips_list; is; is = is->is_next) { 2727 /* 2728 * Look up all the interface names in the state entry. 2729 */ 2730 for (i = 0; i < 4; i++) { 2731 if (ifp == NULL || ifp == is->is_ifp[i]) 2732 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], 2733 is->is_v); 2734 } 2735 } 2736 RWLOCK_EXIT(&ipf_state); 2737} 2738 2739 2740/* ------------------------------------------------------------------------ */ 2741/* Function: fr_delstate */ 2742/* Returns: Nil */ 2743/* Parameters: is(I) - pointer to state structure to delete */ 2744/* why(I) - if not 0, log reason why it was deleted */ 2745/* Write Locks: ipf_state */ 2746/* */ 2747/* Deletes a state entry from the enumerated list as well as the hash table */ 2748/* and timeout queue lists. Make adjustments to hash table statistics and */ 2749/* global counters as required. */ 2750/* ------------------------------------------------------------------------ */ 2751static void fr_delstate(is, why) 2752ipstate_t *is; 2753int why; 2754{ 2755 2756 ASSERT(rw_read_locked(&ipf_state.ipf_lk) == 0); 2757 2758 /* 2759 * Since we want to delete this, remove it from the state table, 2760 * where it can be found & used, first. 2761 */ 2762 if (is->is_pnext != NULL) { 2763 *is->is_pnext = is->is_next; 2764 2765 if (is->is_next != NULL) 2766 is->is_next->is_pnext = is->is_pnext; 2767 2768 is->is_pnext = NULL; 2769 is->is_next = NULL; 2770 } 2771 2772 if (is->is_phnext != NULL) { 2773 *is->is_phnext = is->is_hnext; 2774 if (is->is_hnext != NULL) 2775 is->is_hnext->is_phnext = is->is_phnext; 2776 if (ips_table[is->is_hv] == NULL) 2777 ips_stats.iss_inuse--; 2778 ips_stats.iss_bucketlen[is->is_hv]--; 2779 2780 is->is_phnext = NULL; 2781 is->is_hnext = NULL; 2782 } 2783 2784 /* 2785 * Because ips_stats.iss_wild is a count of entries in the state 2786 * table that have wildcard flags set, only decerement it once 2787 * and do it here. 2788 */ 2789 if (is->is_flags & (SI_WILDP|SI_WILDA)) { 2790 if (!(is->is_flags & SI_CLONED)) { 2791 ATOMIC_DECL(ips_stats.iss_wild); 2792 } 2793 is->is_flags &= ~(SI_WILDP|SI_WILDA); 2794 } 2795 2796 /* 2797 * Next, remove it from the timeout queue it is in. 2798 */ 2799 fr_deletequeueentry(&is->is_sti); 2800 2801 if (is->is_me != NULL) { 2802 *is->is_me = NULL; 2803 is->is_me = NULL; 2804 } 2805 2806 /* 2807 * If it is still in use by something else, do not go any further, 2808 * but note that at this point it is now an orphan. 2809 */ 2810 is->is_ref--; 2811 if (is->is_ref > 0) 2812 return; 2813 2814 if (is->is_tqehead[0] != NULL) { 2815 if (fr_deletetimeoutqueue(is->is_tqehead[0]) == 0) 2816 fr_freetimeoutqueue(is->is_tqehead[0]); 2817 } 2818 if (is->is_tqehead[1] != NULL) { 2819 if (fr_deletetimeoutqueue(is->is_tqehead[1]) == 0) 2820 fr_freetimeoutqueue(is->is_tqehead[1]); 2821 } 2822 2823#ifdef IPFILTER_SYNC 2824 if (is->is_sync) 2825 ipfsync_del(is->is_sync); 2826#endif 2827#ifdef IPFILTER_SCAN 2828 (void) ipsc_detachis(is); 2829#endif 2830 2831 if (ipstate_logging != 0 && why != 0) 2832 ipstate_log(is, why); 2833 2834 if (is->is_p == IPPROTO_TCP) 2835 ips_stats.iss_fin++; 2836 else 2837 ips_stats.iss_expire++; 2838 2839 if (is->is_rule != NULL) { 2840 is->is_rule->fr_statecnt--; 2841 (void)fr_derefrule(&is->is_rule); 2842 } 2843 2844 MUTEX_DESTROY(&is->is_lock); 2845 KFREE(is); 2846 ips_num--; 2847} 2848 2849 2850/* ------------------------------------------------------------------------ */ 2851/* Function: fr_timeoutstate */ 2852/* Returns: Nil */ 2853/* Parameters: Nil */ 2854/* */ 2855/* Slowly expire held state for thingslike UDP and ICMP. The algorithm */ 2856/* used here is to keep the queue sorted with the oldest things at the top */ 2857/* and the youngest at the bottom. So if the top one doesn't need to be */ 2858/* expired then neither will any under it. */ 2859/* ------------------------------------------------------------------------ */ 2860void fr_timeoutstate() 2861{ 2862 ipftq_t *ifq, *ifqnext; 2863 ipftqent_t *tqe, *tqn; 2864 ipstate_t *is; 2865 SPL_INT(s); 2866 2867 SPL_NET(s); 2868 WRITE_ENTER(&ipf_state); 2869 for (ifq = ips_tqtqb; ifq != NULL; ifq = ifq->ifq_next) 2870 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 2871 if (tqe->tqe_die > fr_ticks) 2872 break; 2873 tqn = tqe->tqe_next; 2874 is = tqe->tqe_parent; 2875 fr_delstate(is, ISL_EXPIRE); 2876 } 2877 2878 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 2879 ifqnext = ifq->ifq_next; 2880 2881 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 2882 if (tqe->tqe_die > fr_ticks) 2883 break; 2884 tqn = tqe->tqe_next; 2885 is = tqe->tqe_parent; 2886 fr_delstate(is, ISL_EXPIRE); 2887 } 2888 } 2889 2890 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 2891 ifqnext = ifq->ifq_next; 2892 2893 if (((ifq->ifq_flags & IFQF_DELETE) != 0) && 2894 (ifq->ifq_ref == 0)) { 2895 fr_freetimeoutqueue(ifq); 2896 } 2897 } 2898 2899 if (fr_state_doflush) { 2900 (void) fr_state_flush(2, 0); 2901 fr_state_doflush = 0; 2902 } 2903 2904 RWLOCK_EXIT(&ipf_state); 2905 SPL_X(s); 2906} 2907 2908 2909/* ------------------------------------------------------------------------ */ 2910/* Function: fr_state_flush */ 2911/* Returns: int - 0 == success, -1 == failure */ 2912/* Parameters: Nil */ 2913/* Write Locks: ipf_state */ 2914/* */ 2915/* Flush state tables. Three actions currently defined: */ 2916/* which == 0 : flush all state table entries */ 2917/* which == 1 : flush TCP connections which have started to close but are */ 2918/* stuck for some reason. */ 2919/* which == 2 : flush TCP connections which have been idle for a long time, */ 2920/* starting at > 4 days idle and working back in successive half-*/ 2921/* days to at most 12 hours old. If this fails to free enough */ 2922/* slots then work backwards in half hour slots to 30 minutes. */ 2923/* If that too fails, then work backwards in 30 second intervals */ 2924/* for the last 30 minutes to at worst 30 seconds idle. */ 2925/* ------------------------------------------------------------------------ */ 2926static int fr_state_flush(which, proto) 2927int which, proto; 2928{ 2929 ipftq_t *ifq, *ifqnext; 2930 ipftqent_t *tqe, *tqn; 2931 ipstate_t *is, **isp; 2932 int delete, removed; 2933 long try, maxtick; 2934 u_long interval; 2935 SPL_INT(s); 2936 2937 removed = 0; 2938 2939 SPL_NET(s); 2940 for (isp = &ips_list; ((is = *isp) != NULL); ) { 2941 delete = 0; 2942 2943 if ((proto != 0) && (is->is_v != proto)) { 2944 isp = &is->is_next; 2945 continue; 2946 } 2947 2948 switch (which) 2949 { 2950 case 0 : 2951 delete = 1; 2952 break; 2953 case 1 : 2954 case 2 : 2955 if (is->is_p != IPPROTO_TCP) 2956 break; 2957 if ((is->is_state[0] != IPF_TCPS_ESTABLISHED) || 2958 (is->is_state[1] != IPF_TCPS_ESTABLISHED)) 2959 delete = 1; 2960 break; 2961 } 2962 2963 if (delete) { 2964 fr_delstate(is, ISL_FLUSH); 2965 removed++; 2966 } else 2967 isp = &is->is_next; 2968 } 2969 2970 if (which != 2) { 2971 SPL_X(s); 2972 return removed; 2973 } 2974 2975 /* 2976 * Asked to remove inactive entries because the table is full, try 2977 * again, 3 times, if first attempt failed with a different criteria 2978 * each time. The order tried in must be in decreasing age. 2979 * Another alternative is to implement random drop and drop N entries 2980 * at random until N have been freed up. 2981 */ 2982 if (fr_ticks - ips_last_force_flush < IPF_TTLVAL(5)) 2983 goto force_flush_skipped; 2984 ips_last_force_flush = fr_ticks; 2985 2986 if (fr_ticks > IPF_TTLVAL(43200)) 2987 interval = IPF_TTLVAL(43200); 2988 else if (fr_ticks > IPF_TTLVAL(1800)) 2989 interval = IPF_TTLVAL(1800); 2990 else if (fr_ticks > IPF_TTLVAL(30)) 2991 interval = IPF_TTLVAL(30); 2992 else 2993 interval = IPF_TTLVAL(10); 2994 try = fr_ticks - (fr_ticks - interval); 2995 if (try < 0) 2996 goto force_flush_skipped; 2997 2998 while (removed == 0) { 2999 maxtick = fr_ticks - interval; 3000 if (maxtick < 0) 3001 break; 3002 3003 while (try < maxtick) { 3004 for (ifq = ips_tqtqb; ifq != NULL; 3005 ifq = ifq->ifq_next) { 3006 for (tqn = ifq->ifq_head; 3007 ((tqe = tqn) != NULL); ) { 3008 if (tqe->tqe_die > try) 3009 break; 3010 tqn = tqe->tqe_next; 3011 is = tqe->tqe_parent; 3012 fr_delstate(is, ISL_EXPIRE); 3013 removed++; 3014 } 3015 } 3016 3017 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 3018 ifqnext = ifq->ifq_next; 3019 3020 for (tqn = ifq->ifq_head; 3021 ((tqe = tqn) != NULL); ) { 3022 if (tqe->tqe_die > try) 3023 break; 3024 tqn = tqe->tqe_next; 3025 is = tqe->tqe_parent; 3026 fr_delstate(is, ISL_EXPIRE); 3027 removed++; 3028 } 3029 } 3030 if (try + interval > maxtick) 3031 break; 3032 try += interval; 3033 } 3034 3035 if (removed == 0) { 3036 if (interval == IPF_TTLVAL(43200)) { 3037 interval = IPF_TTLVAL(1800); 3038 } else if (interval == IPF_TTLVAL(1800)) { 3039 interval = IPF_TTLVAL(30); 3040 } else if (interval == IPF_TTLVAL(30)) { 3041 interval = IPF_TTLVAL(10); 3042 } else { 3043 break; 3044 } 3045 } 3046 } 3047force_flush_skipped: 3048 SPL_X(s); 3049 return removed; 3050} 3051 3052 3053 3054/* ------------------------------------------------------------------------ */ 3055/* Function: fr_tcp_age */ 3056/* Returns: int - 1 == state transition made, 0 == no change (rejected) */ 3057/* Parameters: tq(I) - pointer to timeout queue information */ 3058/* fin(I) - pointer to packet information */ 3059/* tqtab(I) - TCP timeout queue table this is in */ 3060/* flags(I) - flags from state/NAT entry */ 3061/* */ 3062/* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */ 3063/* */ 3064/* - (try to) base state transitions on real evidence only, */ 3065/* i.e. packets that are sent and have been received by ipfilter; */ 3066/* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */ 3067/* */ 3068/* - deal with half-closed connections correctly; */ 3069/* */ 3070/* - store the state of the source in state[0] such that ipfstat */ 3071/* displays the state as source/dest instead of dest/source; the calls */ 3072/* to fr_tcp_age have been changed accordingly. */ 3073/* */ 3074/* Internal Parameters: */ 3075/* */ 3076/* state[0] = state of source (host that initiated connection) */ 3077/* state[1] = state of dest (host that accepted the connection) */ 3078/* */ 3079/* dir == 0 : a packet from source to dest */ 3080/* dir == 1 : a packet from dest to source */ 3081/* */ 3082/* Locking: it is assumed that the parent of the tqe structure is locked. */ 3083/* ------------------------------------------------------------------------ */ 3084int fr_tcp_age(tqe, fin, tqtab, flags) 3085ipftqent_t *tqe; 3086fr_info_t *fin; 3087ipftq_t *tqtab; 3088int flags; 3089{ 3090 int dlen, ostate, nstate, rval, dir; 3091 u_char tcpflags; 3092 tcphdr_t *tcp; 3093 3094 tcp = fin->fin_dp; 3095 3096 rval = 0; 3097 dir = fin->fin_rev; 3098 tcpflags = tcp->th_flags; 3099 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2); 3100 3101 if (tcpflags & TH_RST) { 3102 if (!(tcpflags & TH_PUSH) && !dlen) 3103 nstate = IPF_TCPS_CLOSED; 3104 else 3105 nstate = IPF_TCPS_CLOSE_WAIT; 3106 rval = 1; 3107 } else { 3108 ostate = tqe->tqe_state[1 - dir]; 3109 nstate = tqe->tqe_state[dir]; 3110 3111 switch (nstate) 3112 { 3113 case IPF_TCPS_CLOSED: /* 0 */ 3114 if ((tcpflags & TH_OPENING) == TH_OPENING) { 3115 /* 3116 * 'dir' received an S and sends SA in 3117 * response, CLOSED -> SYN_RECEIVED 3118 */ 3119 nstate = IPF_TCPS_SYN_RECEIVED; 3120 rval = 1; 3121 } else if ((tcpflags & TH_OPENING) == TH_SYN) { 3122 /* 'dir' sent S, CLOSED -> SYN_SENT */ 3123 nstate = IPF_TCPS_SYN_SENT; 3124 rval = 1; 3125 } 3126 /* 3127 * the next piece of code makes it possible to get 3128 * already established connections into the state table 3129 * after a restart or reload of the filter rules; this 3130 * does not work when a strict 'flags S keep state' is 3131 * used for tcp connections of course 3132 */ 3133 if (((flags & IS_TCPFSM) == 0) && 3134 ((tcpflags & TH_ACKMASK) == TH_ACK)) { 3135 /* 3136 * we saw an A, guess 'dir' is in ESTABLISHED 3137 * mode 3138 */ 3139 switch (ostate) 3140 { 3141 case IPF_TCPS_CLOSED : 3142 case IPF_TCPS_SYN_RECEIVED : 3143 nstate = IPF_TCPS_HALF_ESTAB; 3144 rval = 1; 3145 break; 3146 case IPF_TCPS_HALF_ESTAB : 3147 case IPF_TCPS_ESTABLISHED : 3148 nstate = IPF_TCPS_ESTABLISHED; 3149 rval = 1; 3150 break; 3151 default : 3152 break; 3153 } 3154 } 3155 /* 3156 * TODO: besides regular ACK packets we can have other 3157 * packets as well; it is yet to be determined how we 3158 * should initialize the states in those cases 3159 */ 3160 break; 3161 3162 case IPF_TCPS_LISTEN: /* 1 */ 3163 /* NOT USED */ 3164 break; 3165 3166 case IPF_TCPS_SYN_SENT: /* 2 */ 3167 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) { 3168 /* 3169 * A retransmitted SYN packet. We do not reset 3170 * the timeout here to fr_tcptimeout because a 3171 * connection connect timeout does not renew 3172 * after every packet that is sent. We need to 3173 * set rval so as to indicate the packet has 3174 * passed the check for its flags being valid 3175 * in the TCP FSM. Setting rval to 2 has the 3176 * result of not resetting the timeout. 3177 */ 3178 rval = 2; 3179 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == 3180 TH_ACK) { 3181 /* 3182 * we see an A from 'dir' which is in SYN_SENT 3183 * state: 'dir' sent an A in response to an SA 3184 * which it received, SYN_SENT -> ESTABLISHED 3185 */ 3186 nstate = IPF_TCPS_ESTABLISHED; 3187 rval = 1; 3188 } else if (tcpflags & TH_FIN) { 3189 /* 3190 * we see an F from 'dir' which is in SYN_SENT 3191 * state and wants to close its side of the 3192 * connection; SYN_SENT -> FIN_WAIT_1 3193 */ 3194 nstate = IPF_TCPS_FIN_WAIT_1; 3195 rval = 1; 3196 } else if ((tcpflags & TH_OPENING) == TH_OPENING) { 3197 /* 3198 * we see an SA from 'dir' which is already in 3199 * SYN_SENT state, this means we have a 3200 * simultaneous open; SYN_SENT -> SYN_RECEIVED 3201 */ 3202 nstate = IPF_TCPS_SYN_RECEIVED; 3203 rval = 1; 3204 } 3205 break; 3206 3207 case IPF_TCPS_SYN_RECEIVED: /* 3 */ 3208 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 3209 /* 3210 * we see an A from 'dir' which was in 3211 * SYN_RECEIVED state so it must now be in 3212 * established state, SYN_RECEIVED -> 3213 * ESTABLISHED 3214 */ 3215 nstate = IPF_TCPS_ESTABLISHED; 3216 rval = 1; 3217 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) == 3218 TH_OPENING) { 3219 /* 3220 * We see an SA from 'dir' which is already in 3221 * SYN_RECEIVED state. 3222 */ 3223 rval = 2; 3224 } else if (tcpflags & TH_FIN) { 3225 /* 3226 * we see an F from 'dir' which is in 3227 * SYN_RECEIVED state and wants to close its 3228 * side of the connection; SYN_RECEIVED -> 3229 * FIN_WAIT_1 3230 */ 3231 nstate = IPF_TCPS_FIN_WAIT_1; 3232 rval = 1; 3233 } 3234 break; 3235 3236 case IPF_TCPS_HALF_ESTAB: /* 4 */ 3237 if (tcpflags & TH_FIN) { 3238 nstate = IPF_TCPS_FIN_WAIT_1; 3239 rval = 1; 3240 } else if ((tcpflags & TH_ACKMASK) == TH_ACK) { 3241 /* 3242 * If we've picked up a connection in mid 3243 * flight, we could be looking at a follow on 3244 * packet from the same direction as the one 3245 * that created this state. Recognise it but 3246 * do not advance the entire connection's 3247 * state. 3248 */ 3249 switch (ostate) 3250 { 3251 case IPF_TCPS_CLOSED : 3252 case IPF_TCPS_SYN_SENT : 3253 case IPF_TCPS_SYN_RECEIVED : 3254 rval = 1; 3255 break; 3256 case IPF_TCPS_HALF_ESTAB : 3257 case IPF_TCPS_ESTABLISHED : 3258 nstate = IPF_TCPS_ESTABLISHED; 3259 rval = 1; 3260 break; 3261 default : 3262 break; 3263 } 3264 } 3265 break; 3266 3267 case IPF_TCPS_ESTABLISHED: /* 5 */ 3268 rval = 1; 3269 if (tcpflags & TH_FIN) { 3270 /* 3271 * 'dir' closed its side of the connection; 3272 * this gives us a half-closed connection; 3273 * ESTABLISHED -> FIN_WAIT_1 3274 */ 3275 nstate = IPF_TCPS_FIN_WAIT_1; 3276 } else if (tcpflags & TH_ACK) { 3277 /* 3278 * an ACK, should we exclude other flags here? 3279 */ 3280 if (ostate == IPF_TCPS_FIN_WAIT_1) { 3281 /* 3282 * We know the other side did an active 3283 * close, so we are ACKing the recvd 3284 * FIN packet (does the window matching 3285 * code guarantee this?) and go into 3286 * CLOSE_WAIT state; this gives us a 3287 * half-closed connection 3288 */ 3289 nstate = IPF_TCPS_CLOSE_WAIT; 3290 } else if (ostate < IPF_TCPS_CLOSE_WAIT) { 3291 /* 3292 * still a fully established 3293 * connection reset timeout 3294 */ 3295 nstate = IPF_TCPS_ESTABLISHED; 3296 } 3297 } 3298 break; 3299 3300 case IPF_TCPS_CLOSE_WAIT: /* 6 */ 3301 rval = 1; 3302 if (tcpflags & TH_FIN) { 3303 /* 3304 * application closed and 'dir' sent a FIN, 3305 * we're now going into LAST_ACK state 3306 */ 3307 nstate = IPF_TCPS_LAST_ACK; 3308 } else { 3309 /* 3310 * we remain in CLOSE_WAIT because the other 3311 * side has closed already and we did not 3312 * close our side yet; reset timeout 3313 */ 3314 nstate = IPF_TCPS_CLOSE_WAIT; 3315 } 3316 break; 3317 3318 case IPF_TCPS_FIN_WAIT_1: /* 7 */ 3319 rval = 1; 3320 if ((tcpflags & TH_ACK) && 3321 ostate > IPF_TCPS_CLOSE_WAIT) { 3322 /* 3323 * if the other side is not active anymore 3324 * it has sent us a FIN packet that we are 3325 * ack'ing now with an ACK; this means both 3326 * sides have now closed the connection and 3327 * we go into TIME_WAIT 3328 */ 3329 /* 3330 * XXX: how do we know we really are ACKing 3331 * the FIN packet here? does the window code 3332 * guarantee that? 3333 */ 3334 nstate = IPF_TCPS_TIME_WAIT; 3335 } else { 3336 /* 3337 * we closed our side of the connection 3338 * already but the other side is still active 3339 * (ESTABLISHED/CLOSE_WAIT); continue with 3340 * this half-closed connection 3341 */ 3342 nstate = IPF_TCPS_FIN_WAIT_1; 3343 } 3344 break; 3345 3346 case IPF_TCPS_CLOSING: /* 8 */ 3347 /* NOT USED */ 3348 break; 3349 3350 case IPF_TCPS_LAST_ACK: /* 9 */ 3351 if (tcpflags & TH_ACK) { 3352 if ((tcpflags & TH_PUSH) || dlen) 3353 /* 3354 * there is still data to be delivered, 3355 * reset timeout 3356 */ 3357 rval = 1; 3358 else 3359 rval = 2; 3360 } 3361 /* 3362 * we cannot detect when we go out of LAST_ACK state to 3363 * CLOSED because that is based on the reception of ACK 3364 * packets; ipfilter can only detect that a packet 3365 * has been sent by a host 3366 */ 3367 break; 3368 3369 case IPF_TCPS_FIN_WAIT_2: /* 10 */ 3370 rval = 1; 3371 if ((tcpflags & TH_OPENING) == TH_OPENING) 3372 nstate = IPF_TCPS_SYN_RECEIVED; 3373 else if (tcpflags & TH_SYN) 3374 nstate = IPF_TCPS_SYN_SENT; 3375 break; 3376 3377 case IPF_TCPS_TIME_WAIT: /* 11 */ 3378 /* we're in 2MSL timeout now */ 3379 rval = 1; 3380 break; 3381 3382 default : 3383#if defined(_KERNEL) 3384# if SOLARIS 3385 cmn_err(CE_NOTE, 3386 "tcp %lx flags %x si %lx nstate %d ostate %d\n", 3387 (u_long)tcp, tcpflags, (u_long)tqe, 3388 nstate, ostate); 3389# else 3390 printf("tcp %lx flags %x si %lx nstate %d ostate %d\n", 3391 (u_long)tcp, tcpflags, (u_long)tqe, 3392 nstate, ostate); 3393# endif 3394#else 3395 abort(); 3396#endif 3397 break; 3398 } 3399 } 3400 3401 /* 3402 * If rval == 2 then do not update the queue position, but treat the 3403 * packet as being ok. 3404 */ 3405 if (rval == 2) 3406 rval = 1; 3407 else if (rval == 1) { 3408 tqe->tqe_state[dir] = nstate; 3409 if ((tqe->tqe_flags & TQE_RULEBASED) == 0) 3410 fr_movequeue(tqe, tqe->tqe_ifq, tqtab + nstate); 3411 } 3412 3413 return rval; 3414} 3415 3416 3417/* ------------------------------------------------------------------------ */ 3418/* Function: ipstate_log */ 3419/* Returns: Nil */ 3420/* Parameters: is(I) - pointer to state structure */ 3421/* type(I) - type of log entry to create */ 3422/* */ 3423/* Creates a state table log entry using the state structure and type info. */ 3424/* passed in. Log packet/byte counts, source/destination address and other */ 3425/* protocol specific information. */ 3426/* ------------------------------------------------------------------------ */ 3427void ipstate_log(is, type) 3428struct ipstate *is; 3429u_int type; 3430{ 3431#ifdef IPFILTER_LOG 3432 struct ipslog ipsl; 3433 size_t sizes[1]; 3434 void *items[1]; 3435 int types[1]; 3436 3437 /* 3438 * Copy information out of the ipstate_t structure and into the 3439 * structure used for logging. 3440 */ 3441 ipsl.isl_type = type; 3442 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0]; 3443 ipsl.isl_bytes[0] = is->is_bytes[0]; 3444 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1]; 3445 ipsl.isl_bytes[1] = is->is_bytes[1]; 3446 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2]; 3447 ipsl.isl_bytes[2] = is->is_bytes[2]; 3448 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3]; 3449 ipsl.isl_bytes[3] = is->is_bytes[3]; 3450 ipsl.isl_src = is->is_src; 3451 ipsl.isl_dst = is->is_dst; 3452 ipsl.isl_p = is->is_p; 3453 ipsl.isl_v = is->is_v; 3454 ipsl.isl_flags = is->is_flags; 3455 ipsl.isl_tag = is->is_tag; 3456 ipsl.isl_rulen = is->is_rulen; 3457 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN); 3458 3459 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 3460 ipsl.isl_sport = is->is_sport; 3461 ipsl.isl_dport = is->is_dport; 3462 if (ipsl.isl_p == IPPROTO_TCP) { 3463 ipsl.isl_state[0] = is->is_state[0]; 3464 ipsl.isl_state[1] = is->is_state[1]; 3465 } 3466 } else if (ipsl.isl_p == IPPROTO_ICMP) { 3467 ipsl.isl_itype = is->is_icmp.ici_type; 3468 } else if (ipsl.isl_p == IPPROTO_ICMPV6) { 3469 ipsl.isl_itype = is->is_icmp.ici_type; 3470 } else { 3471 ipsl.isl_ps.isl_filler[0] = 0; 3472 ipsl.isl_ps.isl_filler[1] = 0; 3473 } 3474 3475 items[0] = &ipsl; 3476 sizes[0] = sizeof(ipsl); 3477 types[0] = 0; 3478 3479 if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1)) { 3480 ATOMIC_INCL(ips_stats.iss_logged); 3481 } else { 3482 ATOMIC_INCL(ips_stats.iss_logfail); 3483 } 3484#endif 3485} 3486 3487 3488#ifdef USE_INET6 3489/* ------------------------------------------------------------------------ */ 3490/* Function: fr_checkicmp6matchingstate */ 3491/* Returns: ipstate_t* - NULL == no match found, */ 3492/* else pointer to matching state entry */ 3493/* Parameters: fin(I) - pointer to packet information */ 3494/* Locks: NULL == no locks, else Read Lock on ipf_state */ 3495/* */ 3496/* If we've got an ICMPv6 error message, using the information stored in */ 3497/* the ICMPv6 packet, look for a matching state table entry. */ 3498/* ------------------------------------------------------------------------ */ 3499static ipstate_t *fr_checkicmp6matchingstate(fin) 3500fr_info_t *fin; 3501{ 3502 struct icmp6_hdr *ic6, *oic; 3503 int type, backward, i; 3504 ipstate_t *is, **isp; 3505 u_short sport, dport; 3506 i6addr_t dst, src; 3507 u_short savelen; 3508 icmpinfo_t *ic; 3509 fr_info_t ofin; 3510 tcphdr_t *tcp; 3511 ip6_t *oip6; 3512 u_char pr; 3513 u_int hv; 3514 3515 /* 3516 * Does it at least have the return (basic) IP header ? 3517 * Is it an actual recognised ICMP error type? 3518 * Only a basic IP header (no options) should be with 3519 * an ICMP error header. 3520 */ 3521 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) || 3522 !(fin->fin_flx & FI_ICMPERR)) 3523 return NULL; 3524 3525 ic6 = fin->fin_dp; 3526 type = ic6->icmp6_type; 3527 3528 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN); 3529 if (fin->fin_plen < sizeof(*oip6)) 3530 return NULL; 3531 3532 bcopy((char *)fin, (char *)&ofin, sizeof(fin)); 3533 ofin.fin_v = 6; 3534 ofin.fin_ifp = fin->fin_ifp; 3535 ofin.fin_out = !fin->fin_out; 3536 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 3537 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 3538 3539 /* 3540 * We make a fin entry to be able to feed it to 3541 * matchsrcdst. Note that not all fields are necessary 3542 * but this is the cleanest way. Note further we fill 3543 * in fin_mp such that if someone uses it we'll get 3544 * a kernel panic. fr_matchsrcdst does not use this. 3545 * 3546 * watch out here, as ip is in host order and oip6 in network 3547 * order. Any change we make must be undone afterwards. 3548 */ 3549 savelen = oip6->ip6_plen; 3550 oip6->ip6_plen = fin->fin_dlen - ICMPERR_ICMPHLEN; 3551 ofin.fin_flx = FI_NOCKSUM; 3552 ofin.fin_ip = (ip_t *)oip6; 3553 ofin.fin_plen = oip6->ip6_plen; 3554 (void) fr_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin); 3555 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 3556 oip6->ip6_plen = savelen; 3557 3558 if (oip6->ip6_nxt == IPPROTO_ICMPV6) { 3559 oic = (struct icmp6_hdr *)(oip6 + 1); 3560 /* 3561 * an ICMP error can only be generated as a result of an 3562 * ICMP query, not as the response on an ICMP error 3563 * 3564 * XXX theoretically ICMP_ECHOREP and the other reply's are 3565 * ICMP query's as well, but adding them here seems strange XXX 3566 */ 3567 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 3568 return NULL; 3569 3570 /* 3571 * perform a lookup of the ICMP packet in the state table 3572 */ 3573 hv = (pr = oip6->ip6_nxt); 3574 src.in6 = oip6->ip6_src; 3575 hv += src.in4.s_addr; 3576 dst.in6 = oip6->ip6_dst; 3577 hv += dst.in4.s_addr; 3578 hv += oic->icmp6_id; 3579 hv += oic->icmp6_seq; 3580 hv = DOUBLE_HASH(hv); 3581 3582 READ_ENTER(&ipf_state); 3583 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 3584 ic = &is->is_icmp; 3585 isp = &is->is_hnext; 3586 if ((is->is_p == pr) && 3587 !(is->is_pass & FR_NOICMPERR) && 3588 (oic->icmp6_id == ic->ici_id) && 3589 (oic->icmp6_seq == ic->ici_seq) && 3590 (is = fr_matchsrcdst(&ofin, is, &src, 3591 &dst, NULL, FI_ICMPCMP))) { 3592 /* 3593 * in the state table ICMP query's are stored 3594 * with the type of the corresponding ICMP 3595 * response. Correct here 3596 */ 3597 if (((ic->ici_type == ICMP6_ECHO_REPLY) && 3598 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 3599 (ic->ici_type - 1 == oic->icmp6_type )) { 3600 ips_stats.iss_hits++; 3601 backward = IP6_NEQ(&is->is_dst, &src); 3602 fin->fin_rev = !backward; 3603 i = (backward << 1) + fin->fin_out; 3604 is->is_icmppkts[i]++; 3605 return is; 3606 } 3607 } 3608 } 3609 RWLOCK_EXIT(&ipf_state); 3610 return NULL; 3611 } 3612 3613 hv = (pr = oip6->ip6_nxt); 3614 src.in6 = oip6->ip6_src; 3615 hv += src.i6[0]; 3616 hv += src.i6[1]; 3617 hv += src.i6[2]; 3618 hv += src.i6[3]; 3619 dst.in6 = oip6->ip6_dst; 3620 hv += dst.i6[0]; 3621 hv += dst.i6[1]; 3622 hv += dst.i6[2]; 3623 hv += dst.i6[3]; 3624 3625 if ((oip6->ip6_nxt == IPPROTO_TCP) || (oip6->ip6_nxt == IPPROTO_UDP)) { 3626 tcp = (tcphdr_t *)(oip6 + 1); 3627 dport = tcp->th_dport; 3628 sport = tcp->th_sport; 3629 hv += dport; 3630 hv += sport; 3631 } else 3632 tcp = NULL; 3633 hv = DOUBLE_HASH(hv); 3634 3635 READ_ENTER(&ipf_state); 3636 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 3637 isp = &is->is_hnext; 3638 /* 3639 * Only allow this icmp though if the 3640 * encapsulated packet was allowed through the 3641 * other way around. Note that the minimal amount 3642 * of info present does not allow for checking against 3643 * tcp internals such as seq and ack numbers. 3644 */ 3645 if ((is->is_p != pr) || (is->is_v != 6) || 3646 (is->is_pass & FR_NOICMPERR)) 3647 continue; 3648 is = fr_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP); 3649 if (is != NULL) { 3650 ips_stats.iss_hits++; 3651 backward = IP6_NEQ(&is->is_dst, &src); 3652 fin->fin_rev = !backward; 3653 i = (backward << 1) + fin->fin_out; 3654 is->is_icmppkts[i]++; 3655 /* 3656 * we deliberately do not touch the timeouts 3657 * for the accompanying state table entry. 3658 * It remains to be seen if that is correct. XXX 3659 */ 3660 return is; 3661 } 3662 } 3663 RWLOCK_EXIT(&ipf_state); 3664 return NULL; 3665} 3666#endif 3667 3668 3669/* ------------------------------------------------------------------------ */ 3670/* Function: fr_sttab_init */ 3671/* Returns: Nil */ 3672/* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 3673/* */ 3674/* Initialise the array of timeout queues for TCP. */ 3675/* ------------------------------------------------------------------------ */ 3676void fr_sttab_init(tqp) 3677ipftq_t *tqp; 3678{ 3679 int i; 3680 3681 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) { 3682 tqp[i].ifq_ttl = 0; 3683 tqp[i].ifq_ref = 1; 3684 tqp[i].ifq_head = NULL; 3685 tqp[i].ifq_tail = &tqp[i].ifq_head; 3686 tqp[i].ifq_next = tqp + i + 1; 3687 MUTEX_INIT(&tqp[i].ifq_lock, "ipftq tcp tab"); 3688 } 3689 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL; 3690 tqp[IPF_TCPS_CLOSED].ifq_ttl = fr_tcpclosed; 3691 tqp[IPF_TCPS_LISTEN].ifq_ttl = fr_tcptimeout; 3692 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = fr_tcptimeout; 3693 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = fr_tcptimeout; 3694 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = fr_tcpidletimeout; 3695 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = fr_tcphalfclosed; 3696 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = fr_tcphalfclosed; 3697 tqp[IPF_TCPS_CLOSING].ifq_ttl = fr_tcptimeout; 3698 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = fr_tcplastack; 3699 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = fr_tcpclosewait; 3700 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = fr_tcptimeout; 3701 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = fr_tcptimeout; 3702} 3703 3704 3705/* ------------------------------------------------------------------------ */ 3706/* Function: fr_sttab_destroy */ 3707/* Returns: Nil */ 3708/* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 3709/* */ 3710/* Do whatever is necessary to "destroy" each of the entries in the array */ 3711/* of timeout queues for TCP. */ 3712/* ------------------------------------------------------------------------ */ 3713void fr_sttab_destroy(tqp) 3714ipftq_t *tqp; 3715{ 3716 int i; 3717 3718 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) 3719 MUTEX_DESTROY(&tqp[i].ifq_lock); 3720} 3721 3722 3723/* ------------------------------------------------------------------------ */ 3724/* Function: fr_statederef */ 3725/* Returns: Nil */ 3726/* Parameters: isp(I) - pointer to pointer to state table entry */ 3727/* */ 3728/* Decrement the reference counter for this state table entry and free it */ 3729/* if there are no more things using it. */ 3730/* */ 3731/* When operating in userland (ipftest), we have no timers to clear a state */ 3732/* entry. Therefore, we make a few simple tests before deleting an entry */ 3733/* outright. We compare states on each side looking for a combination of */ 3734/* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK. Then we factor */ 3735/* in packet direction with the interface list to make sure we don't */ 3736/* prematurely delete an entry on a final inbound packet that's we're also */ 3737/* supposed to route elsewhere. */ 3738/* */ 3739/* Internal parameters: */ 3740/* state[0] = state of source (host that initiated connection) */ 3741/* state[1] = state of dest (host that accepted the connection) */ 3742/* */ 3743/* dir == 0 : a packet from source to dest */ 3744/* dir == 1 : a packet from dest to source */ 3745/* ------------------------------------------------------------------------ */ 3746void fr_statederef(fin, isp) 3747fr_info_t *fin; 3748ipstate_t **isp; 3749{ 3750 ipstate_t *is = *isp; 3751#if 0 3752 int nstate, ostate, dir, eol; 3753 3754 eol = 0; /* End-of-the-line flag. */ 3755 dir = fin->fin_rev; 3756 ostate = is->is_state[1 - dir]; 3757 nstate = is->is_state[dir]; 3758 /* 3759 * Determine whether this packet is local or routed. State entries 3760 * with us as the destination will have an interface list of 3761 * int1,-,-,int1. Entries with us as the origin run as -,int1,int1,-. 3762 */ 3763 if ((fin->fin_p == IPPROTO_TCP) && (fin->fin_out == 0)) { 3764 if ((strcmp(is->is_ifname[0], is->is_ifname[3]) == 0) && 3765 (strcmp(is->is_ifname[1], is->is_ifname[2]) == 0)) { 3766 if ((dir == 0) && 3767 (strcmp(is->is_ifname[1], "-") == 0) && 3768 (strcmp(is->is_ifname[0], "-") != 0)) { 3769 eol = 1; 3770 } else if ((dir == 1) && 3771 (strcmp(is->is_ifname[0], "-") == 0) && 3772 (strcmp(is->is_ifname[1], "-") != 0)) { 3773 eol = 1; 3774 } 3775 } 3776 } 3777#endif 3778 3779 fin = fin; /* LINT */ 3780 is = *isp; 3781 *isp = NULL; 3782 WRITE_ENTER(&ipf_state); 3783 is->is_ref--; 3784 if (is->is_ref == 0) { 3785 is->is_ref++; /* To counter ref-- in fr_delstate() */ 3786 fr_delstate(is, ISL_EXPIRE); 3787#ifndef _KERNEL 3788#if 0 3789 } else if (((fin->fin_out == 1) || (eol == 1)) && 3790 ((ostate == IPF_TCPS_LAST_ACK) && 3791 (nstate == IPF_TCPS_TIME_WAIT))) { 3792 ; 3793#else 3794 } else if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) || 3795 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) { 3796#endif 3797 fr_delstate(is, ISL_ORPHAN); 3798#endif 3799 } 3800 RWLOCK_EXIT(&ipf_state); 3801} 3802 3803 3804/* ------------------------------------------------------------------------ */ 3805/* Function: fr_setstatequeue */ 3806/* Returns: Nil */ 3807/* Parameters: is(I) - pointer to state structure */ 3808/* rev(I) - forward(0) or reverse(1) direction */ 3809/* Locks: ipf_state (read or write) */ 3810/* */ 3811/* Put the state entry on its default queue entry, using rev as a helped in */ 3812/* determining which queue it should be placed on. */ 3813/* ------------------------------------------------------------------------ */ 3814void fr_setstatequeue(is, rev) 3815ipstate_t *is; 3816int rev; 3817{ 3818 ipftq_t *oifq, *nifq; 3819 3820 3821 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) 3822 nifq = is->is_tqehead[rev]; 3823 else 3824 nifq = NULL; 3825 3826 if (nifq == NULL) { 3827 switch (is->is_p) 3828 { 3829#ifdef USE_INET6 3830 case IPPROTO_ICMPV6 : 3831 if (rev == 1) 3832 nifq = &ips_icmpacktq; 3833 else 3834 nifq = &ips_icmptq; 3835 break; 3836#endif 3837 case IPPROTO_ICMP : 3838 if (rev == 1) 3839 nifq = &ips_icmpacktq; 3840 else 3841 nifq = &ips_icmptq; 3842 break; 3843 case IPPROTO_TCP : 3844 nifq = ips_tqtqb + is->is_state[rev]; 3845 break; 3846 3847 case IPPROTO_UDP : 3848 if (rev == 1) 3849 nifq = &ips_udpacktq; 3850 else 3851 nifq = &ips_udptq; 3852 break; 3853 3854 default : 3855 nifq = &ips_iptq; 3856 break; 3857 } 3858 } 3859 3860 oifq = is->is_sti.tqe_ifq; 3861 /* 3862 * If it's currently on a timeout queue, move it from one queue to 3863 * another, else put it on the end of the newly determined queue. 3864 */ 3865 if (oifq != NULL) 3866 fr_movequeue(&is->is_sti, oifq, nifq); 3867 else 3868 fr_queueappend(&is->is_sti, nifq, is); 3869 return; 3870} 3871